<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Lemberg Solutions on Medium]]></title>
        <description><![CDATA[Stories by Lemberg Solutions on Medium]]></description>
        <link>https://medium.com/@WeAreLemberg?source=rss-7925f49ef7c0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*nVIqMq3Iup3WPtMpU2MZIw.jpeg</url>
            <title>Stories by Lemberg Solutions on Medium</title>
            <link>https://medium.com/@WeAreLemberg?source=rss-7925f49ef7c0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 10:53:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@WeAreLemberg/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Top 12 Embedded Software Outsourcing Companies]]></title>
            <link>https://medium.com/@WeAreLemberg/top-12-embedded-software-outsourcing-companies-2d274531a75c?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d274531a75c</guid>
            <category><![CDATA[outsourcing]]></category>
            <category><![CDATA[it-outsourcing-company]]></category>
            <category><![CDATA[embedded-development]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Tue, 19 Aug 2025 12:31:07 GMT</pubDate>
            <atom:updated>2025-08-19T12:31:07.097Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4Rvp-W5z5hj-2bj6VGQjmA.jpeg" /></figure><p><a href="https://lembergsolutions.com/embedded-software-development">Building embedded systems</a> is far from simple, as they require specialized expertise and hands-on engineering skills. And according to a <a href="https://www.mckinsey.com/industries/semiconductors/our-insights/how-semiconductor-companies-can-fill-the-expanding-talent-gap">McKinsey report</a>, the industry is now experiencing a shortage of high-skilled embedded specialists, making in-house hiring especially challenging.</p><p>Therefore, many businesses turn to embedded software services outsourcing as a promising strategy to bridge the talent gap and develop a high-quality embedded solution that meets all relevant standards.</p><p>Read further to explore the best embedded software outsourcing companies and practical tips for selecting the right one for your project.</p><h3><strong>Top 12 Embedded Software Outsourcing Companies</strong></h3><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/cd4efae75541c9986f770b9fbb65a6cf/href">https://medium.com/media/cd4efae75541c9986f770b9fbb65a6cf/href</a></iframe><p>Hourly rate is based on the info provided on the website <a href="http://clutch.co/">clutch.co</a></p><h4><strong>1. </strong><a href="https://lembergsolutions.com/"><strong>Lemberg Solutions</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jIEzxbVzvtdqqNzxATR4QA.jpeg" /></figure><p>Lemberg Solutions is a software and hardware development company that provides end-to-end embedded software outsourcing services to clients around the globe. With a team of more than 200 engineering experts, their expertise also spans Cloud, Data &amp; AI, and Digital. From the discovery phase to the development of a functional product ready for mass production, they assist customers with component selection and hardware and software integration into the solutions.</p><p><strong>Location: </strong>Ukraine (Lviv, Lutsk, Rivne), Germany (Hamburg), Poland (Krakow), United Kingdom (London)</p><p><strong>Focus: </strong><a href="https://lembergsolutions.com/firmware-development">Firmware development</a>, <a href="https://lembergsolutions.com/embedded-linux-development">Embedded Linux</a>, <a href="https://lembergsolutions.com/embedded-ai">Embedded AI</a>, FPGA programming, Hardware development, GUI development</p><p><strong>Industries: </strong><a href="https://lembergsolutions.com/energy">Energy</a>, <a href="https://lembergsolutions.com/healthcare">Healthcare</a>, <a href="https://lembergsolutions.com/consumer-electronics">Consumer Electronics</a>, <a href="https://lembergsolutions.com/automotive">Automotive</a></p><p><strong>Why choose:</strong></p><ul><li>15 years of experience and a team of 120+ skilled embedded software and hardware developers;</li><li>In-house hardware lab;</li><li>Proficiency in embedded C/C++, BLE, low-power firmware, and sensor integration;</li><li>High security &amp; quality management standards;</li><li>Flexible cooperation models;</li><li>End-to-end embedded software development.</li></ul><p><strong>Founded: </strong>2007</p><p><strong>Clients: </strong><a href="https://lembergsolutions.com/case-study/embedded-software-development-cellink">Cellink</a>, <a href="https://lembergsolutions.com/case-study/24x-faster-animal-weighing-computer-vision-based-device">Barkom</a>, <a href="https://lembergsolutions.com/case-study/firmware-development-and-algorithm-integration-fhcs">Freudenberg</a>, <a href="https://lembergsolutions.com/case-study/enhancing-voltfangs-energy-storage-systems-iot-telemetry-solution">Voltfang</a></p><p><strong>Certifications: </strong>STMicroelectronics partner, Digi-Key partner, MotionLab.Berlin partner, ISO 27001:2022, ISO 9001:2015, ISO 13485:2016</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h4><strong>2. </strong><a href="https://yalantis.com/"><strong>Yalantis</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W9L3eeQkiigA3oXxAYNK1w.jpeg" /></figure><p>With 15+ years of experience and over 500 specialists, Yalantis provides custom embedded software development services, building high-quality and real-time embedded systems for IoT devices, sensors, consumer, and industrial electronics. Depending on the customer’s requirements, they deliver embedded development services for OS-based and bare-metal systems.</p><p><strong>Location: </strong>Ukraine (Dnipro), Poland (Warsaw), Estonia (Tallinn), Cyprus (Larnaca)</p><p><strong>Focus: </strong>Embedded architecture design, Firmware development, Embedded AI, Rust consulting, Firmware testing</p><p><strong>Industries: </strong>Healthcare, Industrial, Supply Chain, Automotive, Smart Home &amp; Building, Agriculture</p><p><strong>Why choose:</strong></p><ul><li>Expertise in Rust, C, C++, Zephyr, FreeRTOS, STM32, ESP32, NRF52</li><li>Work with wireless protocols like BLE, Zigbee, and LoRaWAN</li><li>Experience in embedded software development for real-time applications such as remote monitoring</li></ul><p><strong>Founded: </strong>2008</p><p><strong>Clients: </strong>RAK, Bosch, IPG, Toyota Tsusho</p><p><strong>Certifications: </strong>ISO 27701:2019, ISO 9001:2015, ISO 27001:2013, AWS Partner (Advanced Tier Services), AWS Partner (Amazon RDS Delivery)</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h4><strong>3. </strong><a href="https://www.scalosoft.com/"><strong>Scalosoft</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WY1s1lMOZdMRHn_irNyeag.jpeg" /></figure><p>Scalosoft is a provider of tailored software development services with over 80 clients across 13 countries worldwide. They manage the entire embedded software development lifecycle, from concept to delivery, ensuring all hardware components and software are working as intended.</p><p><strong>Location: </strong>Poland (Warsaw, Wrocław, Kraków, Katowice), USA (Austin, TX), Israel</p><p><strong>Focus: </strong>Firmware and driver development, Testing and validation, Optimization and performance tuning, Compliance and certification, IoT integrations</p><p><strong>Industries: </strong>Healthcare, Automotive, Aviation, Renewable Energy, Manufacturing</p><p><strong>Why choose:</strong></p><ul><li>17+ years of experience in custom software development;</li><li>Over 750 projects delivered;</li><li>End-to-end support throughout the entire project lifecycle.</li></ul><p><strong>Founded: </strong>2007</p><p><strong>Clients: </strong>GUKSA, tb.lx, Volvo, Coperion</p><p><strong>Certifications: </strong>Microsoft Gold Partner, ISO 27001</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h4><strong>4. </strong><a href="https://embrox.com/"><strong>Embrox Solutions</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iiT2IKtNwmbXkO2It-JSRw.jpeg" /></figure><p>The company offers full-cycle embedded software development outsourcing services for both hardware and software solutions. The Embrox Solutions team leverages its deep expertise in electrical and mechanical engineering to provide ‌clients with market-ready solutions across healthcare, energy, and media industries.</p><p><strong>Location: </strong>Ukraine (Lviv), Estonia (Tallinn)</p><p><strong>Focus: </strong>Firmware development, Embedded Linux development, Security embedded development, Embedded software consulting</p><p><strong>Industries: </strong>Smart Home, Consumer Goods, Healthcare, Energy, Media &amp; Entertainment</p><p><strong>Why choose:</strong></p><ul><li>Assistance with technical task creation;</li><li>Proficient in low-level programming (C, C++, Rust, Assembly);</li><li>Device prototyping;</li><li>Production assistance and manufacturer sourcing.</li></ul><p><strong>Founded: </strong>2015</p><p><strong>Clients: </strong>SupaChaja, LaylaElectric, Aster MedInnovation, Klimatronik</p><p><strong>Avg. hourly rate: </strong>$25-$49 / hr</p><h4><strong>5. </strong><a href="https://developex.com/"><strong>Developex</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XJIQCA83Z7x-48MSBZYl6A.jpeg" /></figure><p>Over 20 years, Developex has been delivering embedded software outsourcing services, focusing on firmware, middleware, hardware, and software for a wide range of devices. They are trusted by industry-recognized companies for their ability to engineer systems with high performance and low energy consumption rates.</p><p><strong>Location: </strong>Ukraine (Kyiv), Poland (Krakow), Germany (Munich), Canada (Vancouver)</p><p><strong>Focus: </strong>Embedded software, Firmware development, Hardware design consulting, Embedded IoT, Embedded connectivity</p><p><strong>Industries: </strong>Audio &amp; Video,<strong> </strong>Gaming, Smart &amp; IoT, Health, Marine &amp; Automotive</p><p><strong>Why choose:</strong></p><ul><li>20+ years of experience and 400+ delivered projects;</li><li>Different cooperation models;</li><li>Proficient with a wide range of MCUs by TI, STN, NXP, Nordic, Avnera, CSR, Cypress, Atmel, Microchip etc.</li></ul><p><strong>Founded: </strong>2001</p><p><strong>Clients: </strong>Arcade inc, DocBoss, AutoHouse Technologies inc</p><p><strong>Avg. hourly rate: </strong>$25-$49 / hr</p><h4><strong>6. </strong><a href="https://beetroot.co/"><strong>Beetroot</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xBDZvcivzl10_GWXQdlBkw.jpeg" /></figure><p>Beetroot provides team extension services that engage 500+ specialists worldwide. From firmware optimization and real-time systems integration to compliance assurance and custom team training, their specialists ensure top-notch project support at every stage of development.</p><p><strong>Location: </strong>Sweden (Stockholm), Ukraine (Kyiv), Poland (Poznań), Bulgaria (Blagoevgrad)</p><p><strong>Focus: </strong>Embedded system security, Firmware &amp; low-level software development, RTOS development, Hardware-software integration</p><p><strong>Industries: </strong>Healthtech, Greentech, Fintech, Automotive, Industrial Automation &amp; IoT, Consumer Electronics</p><p><strong>Why choose:</strong></p><ul><li>Strong technical expertise in embedded systems;</li><li>Sustainability and ethical business practices;</li><li>Strong focus on security &amp; compliance.</li></ul><p><strong>Founded: </strong>2012</p><p><strong>Clients: </strong>Inspera AS, BranchPattern, Hydrogen Pro, Apex</p><p><strong>Certifications: </strong>ISO 27001, IEC 62443, ISO 26262, GDPR</p><p><strong>Avg. hourly rate: </strong>$25-$49 / hr</p><h4><strong>7. </strong><a href="https://www.globallogic.com/"><strong>GlobalLogic</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kECDunQH-sDj-GnDag29dw.jpeg" /></figure><p>GlobalLogic has been providing software engineering services for over 25 years. Their embedded software development expertise covers hardware and software, operating systems, and communication protocols, ensuring flawless connectivity and optimal performance.</p><p><strong>Location</strong>: Canada, Poland, Ukraine, Romania, Slovakia, Sweden, Israel, UK, USA, Republic of Ireland, Argentina, Croatia, India</p><p><strong>Focus</strong>: Silicon engineering, Embedded software, Hardware design and validation, Connectivity and system integration</p><p><strong>Industries</strong>: Semiconductors, Storage, IoT, Automotive, Telecom and Networking, Consumer Electronics, Healthcare</p><p><strong>Why choose</strong>:</p><ul><li>2,850+ embedded and semiconductor engineers;</li><li>EMI/EMC compliance, environmental validation, and protocol-level verification for interfaces like PCIe and DDRx;</li><li>15+ silicon proven IPs.</li></ul><p><strong>Founded</strong>: 2000</p><p><strong>Clients</strong>: Qualcomm, Cochlear Limited, Pollogen, Harmonic</p><p><strong>Avg. hourly rate</strong>: $100-$149 / hr</p><h4><strong>8. </strong><a href="https://innowise.com/"><strong>Innowise</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VdWfmFj9UNbtFwP7-HbOZw.jpeg" /></figure><p>Innowise is a software development outsourcing company with 2000+ experts across the US and the EU. They specialize in delivering custom-made firmware and software solutions for embedded systems across many industries, including finance, healthcare, and manufacturing.</p><p><strong>Location: </strong>Poland, Lithuania, Italy, Sweden, Norway, France, UK, Germany, Switzerland, Georgia, Uzbekistan, USA, UAE</p><p><strong>Focus: </strong>Embedded software consulting and engineering, Firmware development, Hardware and software integration, Electrical engineering, Embedded Linux development, GUI development</p><p><strong>Industries: </strong>Manufacturing, Consumer Electronics, Automotive, Logistics, Retail, Finance, Defense, Healthcare, Education, Media, Sports, Telecommunication</p><p><strong>Why choose:</strong></p><ul><li>Expertise in wireless technologies such as IrDA, Bluetooth, WiFi, IEEE 802.11, Zigbee, GSM/GPRS;</li><li>Experience in developing control systems, real-time, security, and media streaming products;</li><li>Agile cooperation models.</li></ul><p><strong>Founded: </strong>2007</p><p><strong>Clients: </strong>Talmatic, BeanTech, Netdevops</p><p><strong>Certifications: </strong>ISO/IEC 27001, ISO 9001, Google Cloud Partner, AWS Partner (Advanced Tier Services)</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h4><strong>9. </strong><a href="https://lanars.com/"><strong>LANARS</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fmvHbOeyIfXp3C_jVmnN5A.jpeg" /></figure><p>Hardware and software development company, LANARS, specializes in offering a range of services from IoT solutions and embedded systems to mobile and web apps. Their team leverages in-depth embedded software development expertise to deliver products that meet enterprise-level technical requirements.</p><p><strong>Location: </strong>Norway (Oslo), UK (Edinburgh), Lithuania (Vilnius)</p><p><strong>Focus: </strong>Embedded apps, Firmware, Middleware, Human-machine interfaces (HMIs), Embedded software solutions for industry-grade IoT systems, PCB design</p><p><strong>Industries: </strong>Smart Home, Military, Energy</p><p><strong>Why choose:</strong></p><ul><li>In-house laboratory;</li><li>8+ years of experience and presence in the Nordic region;</li><li>Hardware and software expertise under one roof;</li><li>Professionals with a background in hardware development for the space and aircraft industries.</li></ul><p><strong>Founded: </strong>2016</p><p><strong>Clients: </strong>Liquidtool, Shoplabs, Defigo, Vizium</p><p><strong>Certifications: </strong>ISO 27001</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h4><strong>10. </strong><a href="https://waverleysoftware.com/"><strong>Waverley</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SQowhlbBbI32B0MGiaFvbg.jpeg" /></figure><p>Waverley offers outsourcing services, from custom-made software development to IoT and cloud solutions, for renowned companies and innovative startups. They cover the development of a range of embedded solutions: from boards, middleware, and wearables to IoT devices and ML integration.</p><p><strong>Location</strong>: USA (Palo Alto, CA), Bolivia (Cochabamba), Poland (Wroclaw), Ukraine (Lviv)</p><p><strong>Focus</strong>: Software for Embedded Systems, Embedded Ecosystem Development, Hardware Testing &amp; QA</p><p><strong>Industries</strong>: Fintech, Industrial IoT, Connected Vehicles, Healthtech, Smart Home, Energy</p><p><strong>Why choose</strong>:</p><ul><li>30+ years of experience, being a partner for Fortune 500 companies and cutting-edge startups;</li><li>Full-cycle embedded software development;</li><li>Strong technology expertise in C, C++, multithreading, and TCP/IP.</li></ul><p><strong>Founded</strong>: 1992</p><p><strong>Clients</strong>: Jibo, Spirax, Holu Hou, Seagate</p><p><strong>Certifications</strong>: ISO 27001, AWS Partner</p><p><strong>Avg. hourly rate</strong>: $50-$99 / hr</p><h4><strong>11. </strong><a href="https://newxel.com/"><strong>Newxel</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hD9cFxDK6OVpOHbs6P5eIw.jpeg" /></figure><p>Newxel is a software development outsourcing services provider with a strong presence in over 10 countries around the world. The company’s expertise spans across a bunch of domains like AI &amp; ML, web &amp; mobile, big data, and embedded software development. In embedded software development, their focus lies on Linux Kernel and firmware development.</p><p><strong>Location: </strong>Canada, USA, Poland, Ukraine, UK, Germany, Romania, Portugal, Bulgaria, Israel, Cyprus</p><p><strong>Focus:</strong> Linux Kernel Development, Firmware development</p><p><strong>Industries: </strong>Gaming, Wireless Services, Martech</p><p><strong>Why choose:</strong></p><ul><li>Proficiency in C, C++, and Python;</li><li>Well-versed in the Linux Kernel operating system;</li><li>Hands-on skills in data structure and algorithm concepts, system upgrades, and debugging;</li><li>Conversant with RISC-V instruction set architecture (ISA).</li></ul><p><strong>Founded: </strong>2017</p><p><strong>Clients: </strong>Celeno, Hideez, Firebolt, Feedvisor</p><p><strong>Avg. hourly rate: </strong>$25-$49 / hr</p><h4><strong>12. </strong><a href="https://intechhouse.com/"><strong>IntechHouse</strong></a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rqe8iK7yvf0zsZvcMLE2rA.jpeg" /></figure><p>With over 10 years of experience, IntechHouse specializes in custom embedded solutions development, from low-level device programming to large-scale products. Their team has in-depth knowledge of programming languages such as C/C++ and can effectively adapt the design of embedded systems to product requirements.</p><p><strong>Location: </strong>Poland (Bydgoszcz), USA (Richardson, TX)</p><p><strong>Focus: </strong>Low-level programming, Embedded OS, Embedded IoT solutions, Embedded Software testing, Embedded security</p><p><strong>Industries: </strong>Manufacturing, Healthcare, Logistics, Fintech, Proptech, Oil &amp; Gas</p><p><strong>Why choose:</strong></p><ul><li>Focus on C/C++ and embedded systems design;</li><li>Expertise in smart home application software, industrial equipment, unmanned aerial vehicles, and other embedded systems;</li><li>International quality and security standards.</li></ul><p><strong>Founded: </strong>2003</p><p><strong>Clients: </strong>Pern Group, Mondi, Aviation Technology Solutions Provider</p><p><strong>Certifications: </strong>GPW, ISO 9001, ISO 27001</p><p><strong>Avg. hourly rate: </strong>$50-$99 / hr</p><h3>Checklist for selecting a reliable embedded software outsourcing partner</h3><p>With so many embedded software outsourcing companies on the market, it’s hard to choose the one that truly matches all your requirements. Here is the list of important steps you need to follow to pick a trustworthy vendor:</p><ol><li><strong>Make sure the vendor’s expertise aligns with your needs</strong></li><li><strong>Ensure they have an in-house lab</strong></li><li><strong>Check that all certifications and compliances are intact</strong></li><li><strong>Confirm whether they provide post-delivery manuals and support</strong></li></ol><p>Below, you can discover why these matter for a flawless embedded software development process:</p><h4>Make sure the vendor’s expertise aligns with your needs</h4><p>Choose the vendor whose core services and expertise match your requirements. Not only will the right partner understand your project needs more profoundly, but they will also recommend a more optimal development approach, supporting you in selecting technologies and materials. Check whether your partner has experience working with specific hardware — they must have all the required licenses for IDEs and reliable hardware distributors to get the components for the project.</p><p>For firmware expertise, your potential vendor must have enough experience in designing firmware architectures, identifying the best bare metal technologies; and know how to implement the firmware features you need.</p><h4>Ensure they have an in-house lab</h4><p>An in-house lab is one of the most crucial indicators of an expert embedded company. The lab allows for testing and validating both hardware and software, making sure that the product works flawlessly and as intended. The in-house lab also guarantees that the development and testing of your embedded system are secure, because only the specialists working on your project can access it.</p><h4>Check that all certifications and compliances are intact</h4><p>Your embedded software outsourcing partner must be compliant with all necessary quality and data security standards, holding certifications such as <a href="https://lembergsolutions.com/blog/lemberg-receives-iso-9001-and-iso-27001-certifications">ISO 9001 and ISO 27001</a>. If you want to develop an embedded system within a certain domain, check that your vendor works under industry-specific standards. For example, in <a href="https://lembergsolutions.com/healthcare">medical software development</a>, they must comply with ISO 13485 and IEC 62304, and in the automotive sector, they must follow IATF 16949:2016, ISO/SAE 21434:2021, and ISO 26262 standards.</p><h4>Confirm whether they provide post-delivery manuals and support</h4><p>Your embedded software outsourcing company should provide you with manuals on how to use their solution. It should be understandable — preferably in both written and video formats. Such post-delivery support will make the integration process easier and more efficient. Your employees will have a clear picture of how the solution works; including tips on how to use it to extend its lifespan and what to do in critical situations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MuAu-XGlNeQDw0rra3OAMg.jpeg" /></figure><h3>Why you should invest in embedded software outsourcing development services</h3><p>Discover how investing in a top embedded software development company<strong> </strong>can benefit your project flow:</p><h4><strong>Access to a niche expertise</strong></h4><p>Usually, finding embedded experts locally can be expensive and time-consuming. Hiring embedded outsourcing providers opens a vast opportunity to work with professionals who possess specialized expertise in some programming languages or developing certain embedded systems for your domain at a lower price.</p><h4><strong>More cost-effective</strong></h4><p>If your company only needs a one-time development, there is no need to hire an in-house team, as this will require significant initial costs to onboard them and provide all the necessary equipment. ‌Usually, <strong>embedded development companies</strong> come with their engineering tools and equipment needed for project development, allowing you to save your budget without compromising quality.</p><h4><strong>Better flexibility</strong></h4><p>With an outsourcing partner for embedded development, you can easily manage the size of your team according to your needs. If the scope of the project expands, you can increase the team by requesting more specialists. This way, you can also scale down the team size if the development process does not require as many specialists.</p><h3>FAQ</h3><p><strong>1.Why should I outsource embedded software development instead of hiring in-house?</strong></p><p>Outsourcing embedded software development offers access to a bigger pool of specialists with advanced skills that can be hard to find locally. Outsourcing also saves your budget, as you can find specialists with hourly rates that suit you. These experts usually also have their own tools to develop an embedded system, eliminating the need to invest in additional equipment. Also, you will have better flexibility when it comes to team management, as you can scale up or down your team depending on your needs.</p><p><strong>2. How do I choose a reliable embedded software outsourcing company?</strong></p><p>Before selecting an embedded software partner, make sure to do proper research on all potential candidates. First, check their projects and whether they have already built a similar solution to what you require. After that, examine the cooperation models they propose, and if they have data security and quality management certifications. Only after the research phase is done, request an online meeting to discuss your project requirements and determine whether their development approach suits you.</p><p><strong>3. Which countries are best for outsourcing embedded software development?</strong></p><p>Central and Eastern European countries like Ukraine, Poland, and Romania are considered top destinations for embedded software development outsourcing in terms of cost-quality ratio.</p><p><strong>4. How should IP and data security be handled in embedded outsourcing projects?</strong></p><p>A reliable partner defines the IP ownership at the beginning of the cooperation in the contract, guaranteeing to transfer all the rights to your company. Also, embedded software outsourcing companies must comply with ‌industry-recognized standards like ISO 27001 and ISO 9001.</p><p><strong>5. Can I outsource both hardware and software development together?</strong></p><p>Yes, it is better to outsource both hardware and software development services to one vendor. This approach is recommended because it helps to reduce compatibility and integration issues between the hardware and software.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d274531a75c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top 10 Interesting Startups in IoT You Should Know]]></title>
            <link>https://medium.com/@WeAreLemberg/top-10-interesting-startups-in-iot-you-should-know-e62d6c8c9aa6?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/e62d6c8c9aa6</guid>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Thu, 10 Sep 2020 08:39:30 GMT</pubDate>
            <atom:updated>2020-09-10T10:01:20.304Z</atom:updated>
            <content:encoded><![CDATA[<h4>Ideas to inspire you to create, invent, and act now.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7_fPZjAn3S9Wvsn70LaG0w.png" /></figure><p>Cars, drones, watches, and even the coffee mug we use in the morning… everything can be connected to the <a href="https://lembergsolutions.com/services/iot">Internet of Things</a>. The digitally connected universe of physical devices is not a dream from sci-fi movies anymore. It is a way to make our lives more comfortable and cutting-edge. That’s why lots of companies empower their products with this technology.</p><p>Why follow IoT startups? At the minimum, be ahead of the curve of the latest technological trends, and at the maximum, think about creating your project. Why not? But before fulfilling any of these goals, it is necessary to understand the state of the market and how exactly IoT is helping to translate ideas into successful projects?</p><p><em>Are you looking to build an IoT product? </em><a href="https://lembergsolutions.com/services/iot"><em>Check how Lemberg can help you.</em></a></p><h3>What’s hot on the current state of IoT development?</h3><p>Since the IoT industry is a super hot trend nowadays, countless companies are investing huge amounts of funds and resources. If to define these contributions in figures, then due to <a href="https://www.idc.com/getdoc.jsp?containerId=prAP45362119">the IDC study</a>, organizations worldwide are expected to spend $745 billion on IoT solutions and services this year, with that figure set to exceed $1 trillion in 2022.</p><p>The number of startups is growing, and the ideas they offer are becoming more and more revolutionizing. It implies another feature of the IoT market — the predominance of small startups over tech giants that are more about static positions. Although the business model of startups succeeds in a retrospective, given the high competition in the market, it delivers real benefits to an end-user. As a result, innovations are born.</p><p>So, the reality is, now the small and flexible win.</p><h3>Top 10 in Tech: List of the most promising IoT startups</h3><p>Who is making the most impressive strides in developing IoT? How do they deploy it, and for what purpose? Let’s see.</p><p><strong>#1 Samsara — Logistics facilitator</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gW1aku-LhgLLBj0BGwac_Q.png" /><figcaption>Screenshot from Samsara website</figcaption></figure><p><strong>Founded: </strong>2015<br><strong>Based in: </strong>San Francisco<br><strong>Website:</strong> <a href="https://www.samsara.com/">https://www.samsara.com/</a></p><p><strong>Case in point:<br></strong>In Buddhism, the word samsara means the cycle of death and rebirth to which life in the material world is bound. But, on the market of Internet of Things solutions, Samsara is a fast-growing international company. Having over 700 employees and 5000 available customers throughout the world, Samsara already raised over $ 300 M. What they do: vehicle telematics, driver safety. In other words, this company is striving to make logistics better.</p><p><strong>#2 WINT — Water Intelligence</strong></p><p><strong>Founded: </strong>2011<br><strong>Based in: </strong>New York City<br><strong>Website:</strong> <a href="https://wint.ai/">https://wint.ai/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XPYTa4yAn67ZzMGIXi289A.png" /><figcaption>Screenshot from Water Intelligence website</figcaption></figure><p><strong>Case in point:<br></strong>As far as we are observing a natural resources crisis, we need to think about global environmental issues as soon as possible. The idea was prompted by a team of Israeli enthusiasts who thought about it and developed one of the most successful startups. It performs an important mission for our planet. WINT is successfully developing Water Intelligence, which is a conscious approach to water consumption. Smart sensors allow users to track and regulate water consumption.</p><p><strong>#3 Evrythng — Digitally empowered authentication</strong></p><p><strong>Founded: </strong>2012<br><strong>Based in: </strong>London<br><strong>Website:</strong> <a href="https://evrythng.com/">https://evrythng.com/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JYf7PrZWzqNvZ_y_yx-rOQ.png" /><figcaption>Screenshot from Evrythng website</figcaption></figure><p><strong>Case in point:<br></strong>Another startup that is actively putting its skin in the IoT game. Evrythng is making everything trackable. Based in London, San Francisco, and New York City in 2011, this company presents the SaaS platform to make the entire supply chains reliably visible. This startup helps companies to endow material commodities with sensors, and thus improve brand trust with authentication. Among their customers are top companies like Coca Cola, and Henkel.</p><p><strong>#4 Amiko — Smart healthcare solution</strong></p><p><strong>Founded: </strong>2015<br><strong>Based in: </strong>London<br><strong>Website:</strong> <a href="https://amiko.io/">https://amiko.io/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XGnoumasr9AVFrJq6CdfQw.png" /><figcaption>Screenshot from Amiko website</figcaption></figure><p><strong>Case in point: </strong>The British-Italian startup develops connected inhalers for both individual users and enterprise customers. Respiro is the company’s primary project. The technology consists of three parts: a smart sensor, which can transform any inhaler into a smart connected device, mobile application, and a provider portal. The sensor gathers medication data, analyzes the lung health of a patient, and then sends this information to the cloud.</p><p><strong>#5 Helium — wireless network for cryptocurrency</strong></p><p><strong>Founded: </strong>2013<br><strong>Based in: </strong>San Francisco<br><strong>Website:</strong> <a href="https://www.helium.com/">https://www.helium.com/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PocJr8HEWYOl_1r7_bCZvg.png" /><figcaption>Screenshot from Helium website</figcaption></figure><p><strong>Case in point:<br></strong>Another one ambitious startups in the tech world today. The idea behind this startup is to build a wireless network to earn cryptocurrency through monitoring devices like pet trackers and smart water meters on its web. The company sells $495 IoT hotspots and says its users have deployed 1,201 hotspots across the US so far.</p><p><strong>#6 Axonize — configurable IoT sensors and platform for everyone</strong></p><p><strong>Founded: </strong>2016<br><strong>Based in: </strong>Israel<br><strong>Website:</strong> <a href="https://www.axonize.com/">https://www.axonize.com/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wHPOZ0GHURmKwLHQ9MyzcQ.png" /><figcaption>Screenshot from Axonize website</figcaption></figure><p><strong>Case in point:<br></strong>With an immense focus on innovations, Axonize strives to help other companies to build IoT business faster. For this, the team develops high-quality sensors for all kinds of devices. Moreover, they present The Axonize Platform, which is totally scriptless and allows monitor and analyze amassed data from all devices. This startup quickly evolved into one of the best vendors of IoT services.</p><p><strong>#7 Eyedentify — a watchful eye on drivers</strong></p><p><strong>Founded: </strong>2016<br><strong>Based in: </strong>India<br><strong>Website:</strong> <a href="https://eyedentify.ai/home">https://eyedentify.ai/home</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4mo-2QFs8MyQvsHCgBKkFw.png" /><figcaption>Screenshot from Eyedentify website</figcaption></figure><p><strong>Case in point:<br></strong>Against the backdrop of other IoT startups, this one is aiming at quite an unusual yet important task — enhancing driver and passenger safety while commuting in vehicles. Founded by the group of passionate developers, this company offers hardware &amp; IoT systems leveraging video analytics and deep learning to provide intelligent alerting and reporting in vehicles. Recently they raised $100,000 in funding and are currently looking to raise another $500,000 towards.</p><p><strong>#8 Bastille — the Internet of Radios on a security guard</strong></p><p><strong>Founded: </strong>2014<br><strong>Based in: </strong>San Francisco<br><strong>Website:</strong> <a href="https://www.bastille.net/">https://www.bastille.net/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z5TUtNck6dO2hwWAleBUjw.png" /><figcaption>Screenshot from Bastille website</figcaption></figure><p><strong>Case in point:<br></strong>The company leverages IoT and machine learning algorithms to develop smart radios for threat detection. How do they work? Providing information on every device connected to the system, Bastille highlights what kind of data it is streaming, or if it’s actively attacked in some instances. Such an approach gains particular popularity within corporate environments to ensure security. One more important thing about it is 60 MHz-6 GHz frequency, it covers all available protocols, including Wi-Fi and more.</p><p><strong>#9 Birdie</strong></p><p><strong>Founded:</strong> 2017<br><strong>Based in: </strong>London<br><strong>Website: </strong><a href="https://www.birdie.care/">https://www.birdie.care/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VMILjO2K4qbBORjjTRIviQ.png" /><figcaption>Screenshot from Birdie website</figcaption></figure><p><strong>Case in point: </strong>It helps homecare staff simplify their work. The system clearly presents tasks, reports, and outcomes, with solutions like the Medication Manager, Task Planner, and Family App. Current value: At least £25 million (€28 million).</p><p><strong>#10 Myriota — SpaceX collaborator</strong></p><p><strong>Founded: </strong>2015<br><strong>Based in: </strong>South Australia<br><strong>Website: </strong><a href="https://myriota.com/">https://myriota.com/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Aj6u6zAzIUU-OZ85e2lKtw.png" /><figcaption>Screenshot from Myriota website</figcaption></figure><p><strong>Case in point:<br></strong>he idea of this IoT application is to use of nano-satellite to remove the boundary of getting our IoT data from a distant place and transmit data. The distinctive feature of this startup’s success is its collaboration with SpaceX manufacturer, who regularly launch nano-satellites from Myriota into space.</p><h3>What’s next? Will smart devices change the world?</h3><p><em>“If you think that the internet has changed your life, think again. The Internet of Things is about to change it all over again!” — Brendan O’Brien</em></p><p>While hundreds of incredible Internet of Things inventions have recently hit the market, there is a place for new ideas and projects as well. The companies from this list imply one more reminder: small companies with great ideas can flourish into successful startups. But, only with one condition: you need to have a great business model, implementation, marketing, or targeting. Moreover, development is the crucial stage of any project that can make or break its future. So, think carefully about all these things.</p><p>If you do have worthy IoT ideas, <a href="https://lembergsolutions.com/">Lemberg Solutions</a> can help you bring them to life. We are developing projects of any size and complexity.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e62d6c8c9aa6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Is MVP Development? Why It Is So Important for Startups?]]></title>
            <link>https://medium.com/@WeAreLemberg/what-is-mvp-development-why-it-is-so-important-for-startups-565813171360?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/565813171360</guid>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Wed, 29 Jul 2020 09:58:58 GMT</pubDate>
            <atom:updated>2020-07-29T09:59:28.129Z</atom:updated>
            <content:encoded><![CDATA[<h3>What Is MVP Development? Why Is It So Important for Startups?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h4fSEVNe19vrZytNnISTrA.png" /></figure><p>The time before release is always scary. What will be after? Will customers really need your product? Some applications receive more success, some less. Who knows what results will bring interesting ideas popping in your head? A simple way to know the answer is Minimum Viable Product (MVP) development. If you want to know whether your app is worth pushing it to the market or not, it has all chances to give you a fair answer.</p><p><em>Check Lemberg’s </em><a href="https://lembergsolutions.com/services/startups#mvp-development"><em>MVP development services</em></a><em> for startups.</em></p><h3>Minimum Viable Product: Check If People Need Your Product</h3><p>Your future product will not bring you profits if people do not need it. But if you are passionate about making your product as perfect as it only can, Minimum Viable Product (MVP) development will be a great choice. What is MVP development? The term gradually picked up traction after <em>Eric Ries </em>mentioned it in his book — <em>The Lean Startup (2011).</em> It is a pre-release of a product with a minimum yet enough functionality to satisfy the needs of early customers.</p><blockquote>“A Minimum Viable Product is That version of a new product which allows a team to collect the maximum amount of validates learning about customers with the least effort”. — Eric Ries, The Lean Startup</blockquote><h3>Who Will Benefit From MVP Development?</h3><p>First of all, MVP could be a win-win strategy for small companies and startups that strive to turn an idea into a ready-to-life product. The main feature is the opportunity to validate any idea, receive precise users’ feedback before actual implementation, and so reach goals with fewer risks. That’s why it is often less expensive to launch MVP than release a product and invest money in fixing product failures thereafter.</p><p>Do you know that YouTube originally had a completely different purpose, to be a dating platform? Instagram was also not immediately what we are used to seeing: it was a perplexing application for check-ins with title, think for a minute, Burbn. Now, what is the moral of these stories? Not everyone succeeds right from the start. It is especially true for small startups, which simply cannot invest time and money in a lengthy way of achieving the product-market fit point.</p><p>According to Statista, <a href="https://www.statista.com/statistics/271628/percentage-of-apps-used-once-in-the-us/">25% of apps get abandoned after being used just once</a>. Lots of product owners think they are delivering great products, but customers often have opposite feelings. MVP, in its turn, helps to get out of the atmosphere of extreme uncertainty. It helps to save time, effort, and expensive resources. Moreover, it contributes to learning your customer better, mitigating risks of possible dissatisfaction. Finally, it brings you a working model to pitch to investors.</p><h3>Prototype vs MVP vs Beta Release: What Is Better for Startups?</h3><p>There are several options to try your product in the hands of actual users before the official release: prototype, MVP, and beta release. To make the right choice, let’s get all the ducks in a row and take a closer look at this difference:</p><ul><li><strong>Prototype</strong>. It is the initial stage of <a href="https://lembergsolutions.com/services/startups">product development</a>, where you implement and test general concepts and ideas. A prototype is about big things, without involving more smaller tasks like features creations and more.</li><li><strong>MVP, or the alpha release.</strong> The first version of the product to sell. Since MVP goes from “Minimum Viable Product”, the keyword is MINIMUM. It means that MVP is not about the full realization of features, but only the most important ones.</li><li><strong>Beta release. </strong>Beta is almost the last stage before an official release. The main aim of it is to ensure whether everything is okay with the product in the hands of users under real conditions. As a rule, it is given to a large (but usually selective) group of users.</li></ul><p>An ideal option is to start with a prototype, follow with an MVP, and get to the product. But very often you can be satisfied solely by choosing one option.</p><h3>Build, Measure, Learn: Here Is the Right Approach for MVP</h3><p>One of the biggest challenges many product owners face with Minimum Viable Product is neglecting to stick to necessary steps:</p><ol><li><strong>Find potential users. </strong>To effectively check the results of MVP on functionality, usability, and general perception of your app, it is mandatory to grow the user base. The important point here is to typically find and determine specific users that fit precisely with your ideal customer definition.</li><li><strong>Collect the data. </strong>The further step in the algorithm of the proper start of MVP is the collection of data and measuring feedback delivered by a selected group of potential customers. At this stage, you need to determine the criteria for filtering the information right from the start. Focusing on the right details will give you the right results.</li><li><strong>Extract insights.</strong> After an accurate execution of the previous two steps, there is an equally important task — to analyze all the feedback provided by customers and identify key points that require further refinement. For example, this way, you can find out about the existence of bugs, or vice versa, the lack of the necessary functionality.</li><li><strong>Apply new changes. </strong>The last but no less important step is to take into account all the shortcomings and imperfections of the application and apply updates to reform it. By instantly adding new features and solving old problems, you keep a product up to speed with market requirements, and so enhance it considerably.</li></ol><p>While performing each of these steps, don’t forget about the most important goals: to observe users’ behavior, speak to the early adopters, and make data-driven decisions.</p><h3>What Are the Main MVP Criteria for Success?</h3><p>Implementing too many bells, whistles, and conditions, you are risking to end up going in the wrong direction. To receive expected feedback and achieve the product-market fit point, you need to record all the metrics and insights after MVP. Almost no MVP brings exclusively constructive feedback from customers from the first try, and this is the next task — to grasp the reason for dissatisfaction and to assume what steps can be taken to improve this situation.</p><h3>How to Successfully Launch MVP Product?</h3><p>If you have no prior experience in launching MVP, and the lack of understanding of this sphere, you risk doing it with blunder operating mistakes. But when it comes to startups, every step of development is worth its weight in gold. That’s why even a minor mistake can ruin all the previous work done earlier. Fortunately, there is a way out of these dilemmas.</p><p>Delegating your MVP development to the <a href="https://lembergsolutions.com/services/dedicated-teams">dedicated team</a> of skilled professionals is the most effective way to supercharge your business and, at the same time, play it safe. As a rule, it has already acquired expertise, and so can guarantee for you to obtain desired goals faster without undue stress, risks, and worry. Apart from this, this team can hire and manage other people; they won’t leave you in the middle of the development</p><p>A good sign is when a dedicated team has already overcome many obstacles in terms of various issues and has developed ready-made solutions, for which you would be looking for months. Finally, a dedicated team can save your money, implying the opportunity to estimate your project in advance and calculate your future risks. For this reason, a dedicated team is breeding grounds for MVP. The tricky part is to find a reliable one.</p><p>………………………………………</p><p><em>If you are high on ambition, and passionate about turning innovative ideas into ready-to-life products, </em><a href="https://lembergsolutions.com/"><em>Lemberg Solutions</em></a><em> can help you successfully launch MVP and validate your idea. We follow lean </em><a href="https://lembergsolutions.com/services/startups#mvp-development"><em>MVP product development</em></a><em> and suggest developing solutions that deliver your core value. Feel free to discuss your business and tech needs with us.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=565813171360" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top 10 Machine Learning Applications in Healthcare]]></title>
            <link>https://medium.com/@WeAreLemberg/top-10-machine-learning-applications-in-healthcare-225dcfc0cf0f?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/225dcfc0cf0f</guid>
            <category><![CDATA[ai-in-healthcare]]></category>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ml-application]]></category>
            <category><![CDATA[machine-learning-ai-news]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Tue, 02 Jun 2020 07:57:14 GMT</pubDate>
            <atom:updated>2020-06-02T07:57:14.495Z</atom:updated>
            <content:encoded><![CDATA[<p>The most promising applications up to now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ighz7jB7ha1IRqGj0rLNbQ.png" /></figure><p>Ex-Medica or get ready for a massive Artificial Intelligence tsunami in the medical field. Tracking the patients’ conditions, drug discovery, robotic surgery, faster diagnosis: all these and even more tasks can now be automated thanks to machine learning algorithms. And although everything is just picking up the pace, there are a few noteworthy applications that reshape the face of healthcare. In this article, we provide a market overview and share the most vivid examples of practical ML-powered solutions that give medical staff more time for vital questions.</p><h3>The current state of AI &amp; Machine Learning in medicine</h3><p>ML + medicine is an area where we can observe targeting huge investments. If to define these expenses in numbers, we are heading to reach <a href="https://www.accenture.com/_acnmedia/PDF-49/Accenture-Health-Artificial-Intelligence.pdf#zoom=50">$6.6 billion of investments</a> by 2021.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/760/1*uugGL8kGB6KkiOk-aPQjZg.png" /></figure><p><em>Image credit: </em><a href="https://www.accenture.com/sg-en/insight-artificial-intelligence-healthcare"><em>Accenture</em></a><em>.</em></p><p>Such well-known enterprises like Microsoft, Apple, Google, and IBM stand ahead of the curve of developing medicine with machine learning techniques. Still, apart from this, there are nearly 800 other not so popular yet progressive companies that are engaged in this mission too.</p><p><strong>Why do so many companies are engaged in this? </strong>The answer is obvious. <a href="https://lembergsolutions.com/services/machine-learning">Machine learning</a> is drastically changing clinicians’ roles. ML-powered tools can learn, analyze large amounts of information, and perform better than humans in some visual and auditory recognition tasks. Cutting-edge technologies save time, money, and serve patients more efficiently. For this reason, machine learning systems have all chances to create <a href="https://www.ictandhealth.com/news/accenture-ai-will-lead-to-150-billion-in-annual-savings-by-2026/">$150 billion in annual savings</a> by 2026.</p><h3>Top 10 thrilling use cases of Machine Learning in Medicine</h3><p>What machine learning technologies and mechanisms dominate now in medicine? Let’s explore ten uses cases that have the most significant near-term impact:</p><h3>#1 Natural Language Processing (NLP) for Administrative Tasks</h3><p>As we know, the life of clinical workers is filled with routine daily communications with clients. These responsibilities require constantly updating Electronic Health Records (EHRs), which often consume lots of time and sometimes even cause <a href="https://ehrintelligence.com/news/physician-burnout-is-a-problem-at-83-of-healthcare-organizations?__cf_chl_jschl_tk__=f0993f1b1629892c4de88634aa2ff5bdbe60f9ad-1588948431-0-AWSASKGOlZ0TWX5puadiSo-VULRFcD_DlJIOW9xfUIQrGVr6jwylwQC5FS_Tv3o9APXFmcZR5nJWyuxFzOjmFINCMmj-w5TFwBoSr72_jtVG23k58tBZI4GOW6WdFuUFuPOwZsnAvRRf_uYaW7Xc9qet4QI7A4HcupspIOo6mjMCFYMihL06cDpNgX4bQgjhUUUyL8lZNNDS-zySZb9vNBBo0tjsX75zvQT1ADSfhO9FL73DlmauzxM29OeZXfxAIkJZb35AwgcnzWO04MtEa8_Hek52ZAlhdNq74iZWRjKmJAPGtvUvpLJhhpnfv6suok-P2T8nzbHDQUoEeEpQIOq9-YtT_PLEt_znqLcGitV4">physician burnout</a>. With Natural Language Processing (NLP) and machine learning, it becomes possible to resolve this issue. Being a subfield of AI, NLP deals with everything connected with human-machine interactions through language (speech, written text). For administrative tasks, NLP-based tools classify, extract and summarize the data from clients and direct notes to EHRs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/760/1*uzQR-2b4ebo9QyQifr5dRA.png" /></figure><h3>#2 Detection and classification of brain tumor</h3><p>Radiology is another area where machine learning shows excellent results. According to the <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6268174/">Academic Radiology study</a>, a radiologist needs to review one image every 3–4 seconds to meet workload demands. Deep learning that is a part of ML methods helps to build systems to perform this task faster. The principle behind such technology is typical: a <a href="https://lembergsolutions.com/services/machine-learning">neural network</a> fed by millions of medical X-ray imaging learns to make a final verdict concerning each medical case. The first projects developed in such a way are operating with <a href="https://www.wired.co.uk/article/cancer-risk-ai-mammograms">99% accuracy</a>.</p><h3>#3 Skin Cancer Classification Using Convolutional Neural Networks</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/866/0*xn2DcX3gSuQX1mfN" /></figure><p>One more thrilling ML use case in medicine is the capability to process images. Being a type of Neural Network, CNNs is mainly focused on image recognition and classification. For healthcare, CNNs are beneficial for training machine learning models and make processing thousands of images faster and more efficient. For instance, while dermatologists are making a diagnosis with<a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6231861/"> a 65%-85% accuracy rate</a>, CNNs have <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6562065/">87–95% accuracy</a>.</p><h3>#4 Drug Discovery &amp; Manufacturing</h3><p>Developing a new drug takes time: it is needed nearly 12 years of laboratory testing for drug approval. ML-based systems, in turn, can build new drugs faster than any human team. Such systems are fed by data from all the exciting results, and thus point out at previously unnoticed interaction of molecules. Such an example is <a href="https://arxiv.org/abs/1510.02855">AtomNet</a> software that has already helped create drugs to treat Ebola and multiple sclerosis.</p><h3>#5 Robotic Surgery</h3><p>ML algorithms help to make robotic surgery more sophisticated and allow programs to perform increasingly complex tasks with impossible accuracy for humans. da Vinci X — is one of such projects delivering robot-assisted minimally invasive surgery. And although da Vinci systems do not work independently without doctor’s supervision, they help to carry out many types of complex procedures with more precision, flexibility, and control than is possible with conventional techniques.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*FIuSh_a3A9jOZXJy" /></figure><h3>#6 Deep learning to unlock the potential of Electronic Medical Records</h3><p>Every day the amount of data amassed from EMR becomes not only bigger but more unstructured. It leads to a more complex scenario of data processing. Deep learning techniques, in turn, allow us to automate these processes and help in better <a href="https://arxiv.org/ftp/arxiv/papers/1804/1804.01758.pdf">understanding the evolution of chronic diseases</a> and the prediction of the risk of developing these diseases.</p><h3>#7 Facial recognition identifies rare genetic disorders</h3><p>The face can say a lot about the physical condition of the body, especially in childhood. For example, some genetic diseases can be identified by the shape of the head and the position of the eyes. Of course, this would not be possible without machine learning algorithms and Facial recognition technologies, which allow analyzing large amounts of data and draw certain conclusions in seconds. <a href="https://www.fdna.com/">FDNA</a> has implemented this task as follows. In order to gather a large database of real cases, they created a community platform <a href="https://www.face2gene.com/">Face2Gene</a>, where clinical geneticists downloaded more than 150,000 cases with the identification of rare genetic disorders. After that, it became possible to compare the data obtained from the scan with the existing database and narrow down the disease possibilities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/0*cQwcnDqmF1wdAs-c" /></figure><p><em>Image: FDNA/Nature Medicine</em></p><h3>#8 AI-based tool to recognize a heart attack by phone</h3><p>When such unpredictable incidents with our health like a heart attack occurs, every second is important. But, unfortunately, often calling the 911 services, we don’t even know what’s happening to us. AI system <a href="https://www.corti.ai/">Corti</a>, in turn, can analyze voice messages and recognize whether a person has a heart attack. Through qualitative analysis of NLP, this system recognizes the words and hidden intonations, so it helps medical workers react faster to emergency calls.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/930/0*rTnVOtcX3kWg6vyn" /></figure><h3>#9 Algorithm to diagnose pneumonia</h3><p>Now machine learning can make an improved diagnosis of pneumonia. To detect abnormalities on chest x-rays, developers from the Stanford Machine Learning Group trained a 121-layer Dense Convolutional Network and presented a new algorithm called <a href="https://stanfordmlgroup.github.io/projects/chexnet/">CheXnet</a>. This algorithm can diagnose 14 different pathologies faster than radiologists.</p><h3>#10 Epidemic Outbreak Prediction</h3><p>While the World Health Organization announced the coronavirus epidemic on January 6, <a href="https://bluedot.global/">BlueDot</a> showcased this statement on December 31. The developers of this tool managed to make an early report based on an algorithm that analyzes real-time news reports in the media, official statements by authorities, as well as descriptions of diseases. Another vivid example is <a href="http://www.healthmap.org/en/">HealthMap</a>, a free resource using text-processing algorithms to monitor disease outbreak and provide alerts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*U3pWc5bZ_BYpy8N2" /></figure><p><em>image credit: </em><a href="https://healthmap.org/"><em>HealthMap</em></a></p><h3>Bottom line: The future of AI &amp; ML in Healthcare</h3><p>To err is human. But in some areas, one error makes even bigger ones. For example, a misdiagnosis in medicine can be fatal. But, here is the right time when machine learning systems come to the aid: trained with millions of cases, they can learn, analyze large amounts of information, output valid results and make decisions on their own. And although right now, Medicine + machine learning is still a young direction giving us promises, it is expanding and reaching a point of maturity every day.</p><h3>Looking to develop Machine Learning solutions?</h3><p><a href="https://lembergsolutions.com/">Lemberg Solutions</a> is here to help. Having excellent engineering skills and academic knowledge in data science, machine learning, Internet of Things, we’re part of the industry and build technology that is used worldwide. Applying such <a href="https://lembergsolutions.com/services/machine-learning">machine learning techniques</a> as computer vision, sensor data processing, sound recognition, and NLP, we help our customers to excel with their ideas. Explore our <a href="https://lembergsolutions.com/work/all">case studies</a> and reach out to us anytime through <a href="https://lembergsolutions.com/contact">our website</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=225dcfc0cf0f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Build a Custom Content Recommendation Engine?]]></title>
            <link>https://medium.com/@WeAreLemberg/why-build-a-custom-content-recommendation-engine-be0e5ceec1f0?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/be0e5ceec1f0</guid>
            <category><![CDATA[content-recommendation]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[recommender-systems]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Fri, 26 Jul 2019 14:56:14 GMT</pubDate>
            <atom:updated>2019-07-26T14:56:14.203Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FqFKeTebd7SSMQ0wgGV5WQ.jpeg" /></figure><p>Content recommendation engines help you captivate readers’ interest, <a href="https://www.evergage.com/resources/ebooks/trends-in-personalization-survey-report/">improve clickthrough and conversion rates, and increase revenue from page ads by up to 20%</a>. But deciding whether to purchase an existing content recommendation system — such as Taboola or Outbrain — or build your own can be a real challenge.</p><p>Commercial systems work best for businesses with simple and straightforward goals. If you know that your needs are not going to exceed what paid systems have to offer, there’s no harm in purchasing one.</p><p>However, as your business grows, you might find your commercial solution lacks multilingual support and fine-tuned analytics or doesn’t offer as much control over recommendation accuracy as you might have expected. And although you’ll need a bit of preparation to develop a custom content recommendation engine, it might be well worth it if commercial products don’t satisfy your needs.</p><h3>You should build a custom content recommendation system if…</h3><ul><li>Commercial systems fail to deliver exactly what you’re looking for.</li><li>You want to have control over what your marketing tools can and cannot do.</li></ul><p>Let’s look at a few other cases when custom might be better than ready-made.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IOC5u4ViyIQTuGlLbxyjyQ.jpeg" /></figure><h4>Existing solutions don’t have the features you need</h4><p>Companies behind content recommendation SaaS products strive to reach a wide audience, and creating niche features won’t get them there. If your business is more of an edge case — with goals and approaches different from other businesses in your segment — you won’t benefit from the average feature sets.</p><p>In this case, you might be better off developing a bespoke system with tools that can help your business thrive.</p><h4>You publish for a multilingual audience</h4><p>Although research into natural language processing (NLP) has come a long way, most commercial solutions still only work with content in English. However, if you publish in a different language or have a multilingual platform, chances are you still want to offer accurate content recommendations.</p><p>This was <a href="https://lembergsolutions.com/work/monda-magazin">the case with Monda Magazin</a>, one of our clients. Their online publication is entirely in German, and part of the issue they had with paid content recommendation systems was the inability to accurately analyze German text. To solve this, our team extended and customized existing NLP framework and libraries to build algorithms that can analyze German-language content.</p><p>If you face similar difficulties, building a custom solution is your best bet. This way, you can reap the benefits of smart content recommendations while connecting with your audience in a language they understand.</p><h4>You want a more accurate content recommendation algorithm</h4><p>Commercial content recommendation algorithms might not be accurate enough for your business. Most commercial content recommendation systems implement the statistical twins model, i.e. they group users by shared attributes. Such algorithms find out that a thirty-something-year-old woman from the UK enjoys reading articles about retro car engines, for instance, and then suggest content about retro cars to other thirty-something-year-old women in the UK.</p><p>The statistical twins approach can work well if you publish content about one or two narrow topics to an audience you know well. However, you’ll see a significant drop in accuracy when you begin extending your audience or branch out into new content areas.</p><p>Monda Magazin also had troubles with the accuracy of existing content recommendation algorithms. To solve this challenge, our data science team designed a complex hybrid algorithm that processes article keywords and cross-analyzes them with 36 data points about each Monda reader. With the current state of commercial content recommendation technology, your only way to achieve high recommendation accuracy is to build a custom algorithm.</p><blockquote><em>Read also </em><a href="https://lembergsolutions.com/blog/guide-startups-build-successful-engineering-team"><em>Guide for Startups: Build a Successful Engineering Team</em></a><em>.</em></blockquote><h4>You don’t want third-party products to collect data about your audience</h4><p>Purchasing commercial content recommendation engines comes with a caveat: Third-party products will either collect data about your readers or request access to data you’ve accumulated over time.</p><p>There’s no way around third-party products processing your reader data, as they need data to work. If you don’t want to share data about your audience with third parties, building a custom solution might be your only option.</p><h4>Customizing an existing solution doesn’t work for you</h4><p>While some ready-made content recommendation systems can be customized, the options are often limited. Some systems are architected in a way that doesn’t leave much space for you to customize their features. Or you might find that the size of your business makes customization more expensive than building a solution from scratch.</p><p>In case customizing a paid content recommendation engine is something you can’t or don’t want to do, building your own system is a valid option.</p><p>When considering developing a custom content recommendation system, keep in mind that software development takes time and preparation. Let’s look at a few things you will need to do prior to developing a content recommendation AI.</p><h3>To develop a bespoke content recommendation system, you should…</h3><p>Find the time to focus on development, understand your audience and what you’re trying to achieve, and collect and prepare relevant data on readers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tkWHam0RnEe3cL_0RMAfdQ.jpeg" /></figure><h4>Ensure that your business is in the right place</h4><p>You’ll need time to focus on developing a custom content recommendation AI system. Such a system may take months to build and train properly, and you should take that into consideration when deciding to build one.</p><p>It might be smart to start with a paid content recommendation engine and then move on to building a custom one. You can use a commercial solution while building a custom one even if the recommendation quality of the commercial system isn’t as good as you’d like.</p><h4>Take time to understand your goals and your audience</h4><p>The main attraction of commercial solutions is that you don’t need to think about designing, developing and maintaining software: just buy or subscribe to a product and reap the benefits. Custom development is different — you’re responsible for building a toolkit that you can benefit from.</p><p>To build your own recommendation system, start by studying existing solutions and looking at the typical behavior of your audience. This should help you decide what kinds of features you might need: a tool to monitor audience activity like with Google Analytics, a tool to customize how many articles to suggest, a tool to place different ads on different pages, and so on.</p><p>Once you have a more or less complete list of features, talk to software development companies that have already built content recommendation systems. Most development companies have business analysts to help you define clear product requirements and crystalize your expectations into a concrete product development plan. At Lemberg, Anastasiia Lepuha creates user stories and product requirements. <a href="https://www.linkedin.com/feed/update/urn:li:activity:6550043638955679744/">You can learn more about Anastasiia in one of our LinkedIn posts.</a></p><h4>Start collecting and preparing relevant data about your readers</h4><p>Smart solutions require quite a bit of quality training data. If you have a data engineer on your team, start collecting and preparing data on your readers as soon as you can: the more data you give to your data scientist, the better the algorithm they’ll be able to design for you.</p><blockquote><em>You can learn more about collecting and preparing training data for AI in our article </em><a href="https://lembergsolutions.com/blog/how-prepare-training-data-better-ai"><em>How to Prepare Training Data for Better AI?</em></a></blockquote><h3>Summary</h3><ul><li>Content recommendation engines can offer a 20%+ boost to revenue from page ads as well as to your clickthrough and conversion rates.</li><li>Commercial content recommendation engines work best for businesses with simple, straightforward goals.</li><li>If you find commercial solutions lacking, you can build a custom solution that perfectly meets your needs.</li><li>To build a quality recommendation system, prepare thoroughly: allocate enough resources, analyze existing solutions, understand your audience, and collect relevant data.</li><li>Developing a custom content recommendation AI isn’t easy, but it might be worth the effort if existing solutions don’t help you achieve your goals.</li><li>Learn about the custom content recommendation engine we built at Lemberg</li><li>To decide whether to build your own custom recommendation system or purchase a ready-made system, consider your goals. If existing products aren’t good enough to help you reach those goals, develop your own software that does exactly what you need.</li></ul><p><a href="https://lembergsolutions.com/contact">Write a quick note to Roy Vikovych</a> and ask him about the bespoke content recommendation system we developed for Monda Magazin.</p><p><em>Originally published at </em><a href="https://lembergsolutions.com/blog/why-build-custom-content-recommendation-engine"><em>https://lembergsolutions.co</em></a><a href="https://lembergsolutions.com/blog/why-build-custom-content-recommendation-engine"><em>m/blog/why-build-custom-content-recommendation-engine</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be0e5ceec1f0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Guide for Startups: Build a Successful Engineering Team]]></title>
            <link>https://medium.com/@WeAreLemberg/guide-for-startups-build-a-successful-engineering-team-19750f5667c7?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/19750f5667c7</guid>
            <category><![CDATA[dedicated-team]]></category>
            <category><![CDATA[startup-lessons]]></category>
            <category><![CDATA[startup-hiring]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Fri, 17 May 2019 06:17:07 GMT</pubDate>
            <atom:updated>2019-05-17T06:17:07.451Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*onR6Jvm5cTblc7-Kmw8SwA.jpeg" /></figure><p>Hiring competent software engineers is a real problem for many startups. You can have the best idea in the world along with all the funding you need, but make a couple of wrong hires and it all goes down the drain.</p><p>While you can always <a href="https://lembergsolutions.com/services/dedicated-teams">hire a reliable dedicated engineering team</a>, it’s wise to maintain a healthy balance between in-house and dedicated specialists. Based on our <a href="https://lembergsolutions.com/work/cell-garden">experience working with startups</a>, we’ll share our advice on hiring in-house and dedicated engineers for your startup: where to begin, how to know what dedicated developers you need, and how to know whether developers are truly professionals.</p><h3>First, get some managing partners</h3><p>Many founders don’t have enough engineering and recruiting experience to hire developers for a startup. So don’t start with that!</p><p>Start by finding experienced partners who can help you hire engineers and manage product development. Building a tech startup is probably unlike anything you’ve done before, and having the right people by your side from the start will help you build a truly worthwhile team.</p><h4>Chief technology officer (CTO)</h4><p>The CTO is your go-to technical partner. This is the person who should be making major decisions about the course of product development, maintaining the system architecture, interviewing engineers, and auditing the engineering team’s work.</p><p>Hiring a reliable CTO is far from easy: Your CTO should be someone whose opinion and expertise you can fully trust. If you’re not sure you can find a suitable professional yourself, consider hiring a CTO as a service. This is a relatively new commodity on the market: Companies lend their own CTOs to smaller businesses to provide the help they need. A CTO acquired this way can be a member of your team for as long as you need them — until your product is fully on the market or you’ve found an in-house CTO, for instance.</p><h4>Project manager (PM)</h4><p>A project manager oversees the development of your product. They encourage the engineers to maintain technical documentation, organize and manage the development process, communicate directly with developers, you, and all other stakeholders, and handle a whole variety of other tasks.</p><p>You should be able to trust your PM. They must understand and respect your needs while guiding your project through all of the inevitable hurdles. So try to get an in-house PM: they’ll be the person who keeps a steady hand on your project’s pulse, and you want to be able to talk to them in person at any time.</p><p>However, you can also hire a dedicated PM — and a really good one, too. Just look for companies that provide project managers as part of their dedicated teams.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oqPsQJKSgBMVa4GUWsx9pw.jpeg" /></figure><h3>Next, prepare the initial product documentation</h3><p>The initial product documentation consists of requirements, user stories, acceptance criteria, and other materials that describe your product very clearly. And when we say very, we mean it: Well-written documentation leaves no space for interpretation. You should be able to tell exactly which button does what and which screen takes the user where.</p><p>With product documentation, you’ll</p><ul><li>understand what specialists your product needs and when</li><li>know exactly what’s being developed when</li><li>make sure your product is developed correctly.</li></ul><p>Creating and maintaining product documentation is likely the most important responsibility for a startup. The more meticulous you are at this, the easier it will be for you to find developers and make sure that your product gets developed. In fact, even if you have to put the entire startup on hold for a while, well-written documentation will enable you to resume development at any point.</p><p>Some project managers take on the task of creating and maintaining product documentation. Other times, PMs only agree to maintain documents and want to see the initial requirements before taking the job. In that case, look for a business analyst (BA). They’ll take you through several sessions of rigorous interviews to determine your exact expectations for the final product and compose product requirements, user stories, and acceptance criteria based on those expectations. You don’t, however, need an in-house BA, and you might be be better off with a dedicated one: their services are invaluable at the very beginning, but a skilled PM can easily cover the rest.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TtslhQfx54Q6ZvRmL_N71g.jpeg" /></figure><p><em>Read also: </em><a href="https://lembergsolutions.com/blog/discovery-phase-one-small-step-you-one-giant-leap-your-project-success"><em>Discovery Phase: One Small Step for You, One Giant Leap for Your Project Success</em></a></p><h3>Proceed by setting a project development timeline</h3><p>A project development timeline visualizes how long it will take to develop each component of the product. To create a timeline, a project manager takes the product requirements (which describe the product’s features) and divides them into sizeable chunks. The exact division will depend on the PM’s management approach. But no matter how you slice it, a project will always be easier to tackle if you divide it into small parts.</p><p>We asked Anastasiia Sirenko, a Project Manager at Lemberg, to create an example project development timeline. Her timeline outlines the first phase of developing a mobile application that consists of a backend API and two native mobile apps, one for iOS and one for Android.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AYwysqo7tW-hCkGm" /></figure><p>Given that the customer has provided us with a finalized user interface and user experience design (UI/UX), we would dedicate a team of five to deliver the finished product in about two months. The dedicated team would consist of:<em> 1 full-time backend developer, 1 full-time Android developer, 1 full-time iOS developer, 1 part-time quality assurance specialist, 1 part-time project manager.</em></p><p>See what we did there? We took the project requirements, divided them into phases, quickly pointed out that a project of this scope will require five people to develop, and estimated the amount of time their work will take. With a project development timeline, you should be able to do the same for your own project.</p><h3>Now, decide who to hire in-house</h3><p>This is definitely a tricky decision. Depending on your location, hiring in-house developers can be very expensive, and their rates won’t necessarily correspond with their professionalism.</p><p>If your budget is tight and you’re not sure whether your company can sustain an in-house developer, don’t hire one yet. Start by getting a qualified dedicated team and slowly work your way up to building an in-house team. Learn more about the development process, speak to your dedicated developers, and regularly talk to your PM and CTO. This approach will ensure that you don’t waste your money on hiring people you’re not completely sure you need.</p><p>If you do have the opportunity to hire an in-house developer, start with a backend developer. On most projects, backend developers are the ones most familiar with the system structure. They’ll be able to understand your product’s architecture and share it with other developers, whether dedicated or in-house.</p><p>Then, depending on your product’s needs — which you can learn from your CTO and PM — start forming an in-house engineering team. It can consist of: <em>UI/UX designers, Solution architects, Back-end developers, Front-end developers, Firmware developers, Hardware developers, iOS developers, Android developers, Cross-platform mobile developers, Data scientists, QA specialists, DevOps.</em></p><p>If you want to be super secure in preserving the knowledge of your product’s system structure, hire two similarly experienced backend developers to share responsibilities. This way, they’ll be able to back each other up in case of emergency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*izJA7GsS4KL0XLXzdBSgLg.jpeg" /></figure><p><em>Read also: </em><a href="https://lembergsolutions.com/blog/guide-managing-your-remote-product-development-team-clients-handbook"><em>Guide to Managing Your Remote Product Development Team [Client’s Handbook]</em></a></p><h3>Always make sure to get real professionals</h3><p>Each startup employee has a huge influence on the company’s success. That’s why you should make sure you hire professional engineers with proven skills.</p><p>Ideally, there should be at least two people interviewing each new engineer: your CTO (or an in-house developer if you have one) and your PM. The CTO will make sure that the developer knows their stuff, while the PM will see that they fit with the rest of the team. However, even though teamwork is beyond important in software and hardware development, this doesn’t necessarily mean that all engineers should be open and easygoing.</p><p>Although there are many articles that define good software engineers based on their communication skills, proactivity, supportiveness, and ability to propose amazing ideas, the typical real-life developer is far from this. What defines a great developer is:</p><ul><li>technical expertise</li><li>responsibility for their work</li><li>respect for their colleagues.</li></ul><p>A great developer should be able to do what the PM asks them to and do it well. Otherwise, your project will struggle to be successful. These are the people you should really be looking for, not just someone who’s gotten really good at selling themselves during interviews.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jHHL9E2x9tCYaAYj7uQ7lA.jpeg" /></figure><h3>Finally, find your dedicated engineers</h3><p><a href="https://lembergsolutions.com/work/all">Lemberg has a lot of experience working with both startups and established businesses.</a> We mentor at many European startup accelerators. Startups are the future, and we want to help them thrive.</p><p>If you have a startup idea and are looking for a professional development team, <a href="https://lembergsolutions.com/contact">send us a quick message</a>. We could be the ones to help you turn your idea into reality!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19750f5667c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CES 2018: Lemberg’s Insights]]></title>
            <link>https://medium.com/@WeAreLemberg/ces-2018-lembergs-insights-7f8a9908f7c8?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/7f8a9908f7c8</guid>
            <category><![CDATA[iot]]></category>
            <category><![CDATA[wearables]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[ces]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Tue, 23 Jan 2018 11:44:20 GMT</pubDate>
            <atom:updated>2018-01-23T11:44:20.064Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*VQE3vaK9e4nsW81-." /></figure><p><em>Originally published at </em><a href="http://blog.lemberg.co.uk/agile-project-management-through-looking-glass"><strong><em>blog.lemberg.co.uk</em></strong></a><strong><em>.</em></strong></p><p><em>By </em><a href="http://blog.lemberg.co.uk/blog/slavic-voitovych"><em>Slavic Voitovych</em></a></p><p>Now that we are back from fabulous Las Vegas show of the year — <a href="https://twitter.com/hashtag/CES2018?src=hash">#CES2018</a> — we can take a moment to reflect on what this event was all about. Of course it was about the most recent technology, but let’s see what was actually hot. Our focus was mainly on automotive, robotics, smart home, pet trackers, and other impressive consumer electronics.</p><h3>Voice Assistants: Google wants it all</h3><p>Right from the beginning we realized that voice assistants will be the dominating subject. Google has never got as much exposure as it was getting here at CES Las Vegas and also earlier at <a href="http://blog.lemberg.co.uk/mwc17-lembergs-insights-IoT-conference">MWC Barcelona</a> event. The reason is that Google started feeling uncomfortable as many smart products fueled by Alexa started propelling Amazon’s business. And here at CES the Hey Google ads were all over Vegas and inside CES, not only on Google’s booth, but also on a big number of startup booths to showcase how Google Assistant can help you get more out of a particular hardware product. For example, a popular Tile beacon that can help you find your tagged belongings. You could ask Google Assistant to find your keys with a Tile on it, pretty much the same use case with a lost backpack, ask Google Assistant and it will tell you the last coordinates of your lost backpack.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*3KW4LrjyGbr1Q8qt.jpg" /></figure><p>Images source: <a href="https://www.cnbc.com/2018/01/08/google-announces-smart-displays-actions-auto-integration-for-assistant.html">https://www.cnbc.com/2018/01/08/google-announces-smart-displays-actions-auto-integration-for-assistant.html</a></p><p>Google took even a bigger step forward and partnered up with hardware manufacturers like JBL, Lenovo, LG and Sony to build “smart displays” powered by Google Assistant in a war against Amazon. This company really wants to have their powerful AI bot in all of your gadgets.</p><h3>Automotive and future mobility experiences</h3><p>Of course the voice assistants battle is pretty hot in automotive business too. This time Daimler has announced something even more interesting — <a href="https://www.daimler.com/innovation/case/connectivity/mbux-2.html">MBUX smart multimedia system</a> and own in-car voice activated assistant. On the hardware side it’s powered by Nvidia chips and your infotainment UX feels really smooth, just like an iPhone experience. Finally your infotainment system will be not just OK in terms of usability, you are gonna love it! And from now on you can also talk to your car by saying “Hey Mercedes”. It will read natural commands like “I’m cold” and initiate your preferred climate control settings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*IR_j-qYWy-mPDliG.jpg" /></figure><p><em>Images source: </em><a href="https://www.mercedes-benz.com/en/mercedes-benz/innovation/mbux-mercedes-benz-user-experience-revolution-in-the-cockpit/"><em>https://www.mercedes-benz.com/en/mercedes-benz/innovation/mbux-mercedes-benz-user-experience-revolution-in-the-cockpit/</em></a></p><p>We couldn’t stand but noticed that some startups are showing how you can develop your own voice assistant that can fit each and every car, even if it’s an older make you are driving, not a brand new electric vehicle with all the bells and whistles. Meet <a href="https://www.hellochris.ai/">Chris </a>by German Autolabs — your digital co-driver with voice and gesture control technology, alongside artificial intelligence (AI). The combination of an interoperable, scalable software platform for cognitive assistance with a retrofit hardware device democratizes the latest connected car technology — making it accessible to everyone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*Mowdmu8Iy_7tXiCP.jpg" /></figure><p><em>Images source: </em><a href="https://twitter.com/germanautolabs/status/950785637904338944">https://twitter.com/germanautolabs/status/950785637904338944</a></p><p>A lot of hype was created by the newly introduced <a href="http://www.byton.com/">Byton concept car </a>— an all-electric SUV with a driver tablet on the steering wheel and abnormally big display on the dashboard, it’s just super big! The front seats rotate 12 degrees to give those in the back a better view on the big screen. This is just a cinema on the wheels. The company calls its interior a Digital Lounge — well it really is what they say. They’ve also got Alexa inside, gesture recognition, facial recognition, Level 3 autonomous drive right from production, it’s 5G-ready, and even reads your body data to give you useful healthcare insights.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*Yu5riXalMijKhVrw.jpg" /></figure><p>We have checked out Nissan’s electric <a href="https://www.nissan.co.uk/experience-nissan/concept-cars/nissan-imx.html">IMx concept car</a>. Its aggressive styling and visual appeal would make you say Whoa! Definitely an attractive car with stunning interior and ability to engage Nissan’s self-driving ProPILOT system. Nissan is also working to bring a smart assistant into the car, but they don’t disclose which partners they will work with. But the idea is to have a smart assistant follow the driver even outside the car. This is a concept car but it represents Nissan’s general direction towards “Intelligent Mobility”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*qoInkyp1p78lrfhU.jpg" /></figure><p>Toyota has introduced its <a href="https://www.cnet.com/roadshow/news/toyota-thinks-this-self-driving-box-is-the-future-of-transport/">e-Pallete</a> — a self-driving minibus that can give you a ride or can be used for retail deliveries. The most peculiar thing was that e-Pallete is more of a platform than just a hardware product. Akio Toyoda, the company’s president, said that cars will become extensions of our phones and computers, it’s all about the platform not hardware or software. That’s why companies which keep him up at night are not Ford, Nissan, or Daimler… it’s Google, Apple and Facebook.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*q_iDugFdOP_27QnH.jpg" /></figure><p><em>Images source: </em><a href="https://www.cnet.com/roadshow/news/toyota-thinks-this-self-driving-box-is-the-future-of-transport/">https://www.cnet.com/roadshow/news/toyota-thinks-this-self-driving-box-is-the-future-of-transport/</a></p><p>Another amazing company called <a href="https://navya.tech/">Navya</a> showcased its 100% autonomous, driverless and electric transport solutions for the first and last mile. They have deployed 60 vehicles worldwide and their typical car is just full of sensors — 6 cameras, 4 odometers, 10 LIDARs, 2 GNSS antennas, 4 radars, 1 IMU, V2X, 4G. There is no steering wheel, no pedals, nobody can drive it so it’s completely self-driving. During a demo we have been reassured that it can drive even in heavy rain conditions. Haven’t got a robo-taxi in your city yet? Be sure to get one soon.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*QUM_UGCyB7Qju1hh.jpg" /></figure><p><a href="https://global.yamaha-motor.com/showroom/event/tokyo-motorshow-2017/exhibitionmodels/motoroid/">Motoroid </a>from Yamaha looked like a futuristic motorcycle from a movie. It has very cool features like auto-balance, facial and gesture recognition based on artificial intelligence (AI). The wings on the back are a special way to support the rider’s back at high speeds and also connect with the bike through haptic feedback.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*3edHpaE35DDJM89Z.jpg" /></figure><h3>Robots for your home, pets and sleep</h3><p>We were impressed with a cute family robot called Buddy by Blue Frogs Robotics. With all the features that <a href="http://www.bluefrogrobotics.com/en/buddy/">Buddy </a>has to offer it can very well become a valuable member of your household. Buddy can inform you of personal important events, wake you up in the morning, control your smart home via voice commands, send and respond to messages or calls, it will also watch over your home while you are away, and so much more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*u1-sXGAGN-_oQDcL.jpg" /></figure><p>Pet tracking devices by <a href="https://www.jagger-lewis.com/">Jagger &amp; Lewis</a> do not actually track just the activity of your animal, but they rather track your pet’s emotions and behaviors. This is pretty unique and can help pet owners be closer emotionally to their beloved animals.</p><p><a href="https://www.adoptlaika.com/">Laïka </a>is an interactive companion for you &amp; your dog. This robot can either work autonomously and engage with you pet and also reward it. Or you can control it via your smartphone remotely and don’t let your dog feel lonely when you are not home. Great way to stay connected with your pet and keep your dog away from chewing the furniture in your house.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*p28a28dcNdOMyxt6.jpg" /></figure><p><a href="https://www.somnox.nl/">Somnox</a> is building a sleep robot you have never experienced before. It’s truly cute and leaves a special feeling about it once you try it yourself. We got a little addicted actually while testing it and just had a feeling we want this experience again. Once you hold it tight to your chest it starts moving its middle part almost like it’s breathing and could also give you a nice sound of a heartbeat to calm you down just before you fall asleep. So it feels like holding a little animal to yourself that would give you a positive and relaxing emotion.</p><p>Nokia introduced its health product called <a href="https://health.nokia.com/uk/en/sleep">Nokia Sleep</a>. Basically it’s a Wi-Fi connected mattress pad that tracks your sleep patterns, or even snoring, if you do that kind of thing. It integrates with IFTTT so you can do amazing things with it. For example, based on your sleeping pattern the system can dim your smart lights or completely shut them off before you fall asleep. Another use case is you could automatically lower the temperature in your house when you fall asleep to keep your body healthier at night.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*Mh-Lqg3Vy_ySyn_b.jpg" /></figure><h3>BioTech</h3><p>Something that attracted our attention was an innovative Swiss product called <a href="http://www.biowatch.ch/">BioWatch</a>. This is a biometric sensor built into your wristwatch buckle that can scan the unique pattern of your wrist veins. This is how it will know that it’s you. And from there you can do whatever you want: access doors, pay for your groceries, access your car, and much more. Nobody can get advantage of it because it works only on your wrist. We look forward to seeing this technology widely used in masses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*9tJCK9cJsRauKl4b.jpg" /></figure><p>All in all CES2018 went pretty well. Undoubtedly artificial intelligence (AI) is the focal point because it powers today’s and tomorrow’s technology: smart homes, connected cars, all of the <a href="https://lemberg.co.uk/services/iot-development-services">Internet of Things</a> (IoT). We saw a lot of new and old “smart” gadgets here, but it remains clear that software will define how cool &amp; smart your product may be, to make the user say and feel the Whoa! effect. And if you are like me and failed to take the overbooked autonomous-driving Lyft this year, there is always next year!</p><p>Feel free to share your feedback below. Drop us some notes about what impressed you at CES2018?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7f8a9908f7c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Motion gesture detection using Tensorflow on Android]]></title>
            <link>https://medium.com/@WeAreLemberg/motion-gesture-detection-using-tensorflow-on-android-9670249b8e9e?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/9670249b8e9e</guid>
            <category><![CDATA[gesture-recognition]]></category>
            <category><![CDATA[tensorflow]]></category>
            <category><![CDATA[android]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Mon, 25 Dec 2017 08:09:11 GMT</pubDate>
            <atom:updated>2017-12-25T08:09:11.469Z</atom:updated>
            <content:encoded><![CDATA[<p><em>By </em><a href="http://blog.lemberg.co.uk/blog/ruslan-yanchyshyn"><em>Ruslan Yanchyshyn</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/950/0*tsfDLiWTF2rDKX5f." /></figure><h3>Introduction</h3><p>There are many different ways to interact with smartphones these days: touchscreen, hardware buttons, fingerprint sensor, video camera (e.g., face recognition), D-PAD, handset controls, and so on. But how about using motion gestures?</p><p>For example, a quick move of the phone to the right or left while holding it in the hand can very precisely indicate the intention to switch to the next or previous song on a playlist. Or you might quickly flip the phone down and then up to refresh the application content. Introducing such interaction looks very promising and adds a new dimension to the UX — literally. This article will describe how to achieve this goal using machine learning and the Tensorflow library on Android.</p><h3>Description</h3><p>Lets clarify our goal. We would like for the phone to be able to recognize quick motions to the left and right.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*ySvRI1-pbAaLNdc2.jpg" /></figure><p>We would like the implementation to reside in a separate Android library so that it can be easily integrated into any other application.</p><p>The motions can be captured on the phone with several sensors: accelerometer, gyroscope, magnetometer, etc. Later, the batch of these motions can be used in machine learning algorithms for training and subsequent recognition.</p><p>To capture the data, an Android application will be developed. Preprocessing and training will be performed on a PC in a Jupyter Notebook environment using Python and the TensorFlow library. Gesture recognition will be implemented in a demo Android application with resulting training data. Finally, we will develop a ready-to-use Android library for gesture recognition that can be easily integrated into other applications.</p><p>Here is the high-level plan for implementation:</p><ol><li>Collect data on the phone</li><li>Design and train the neural network</li><li>Export the neural network to the phone</li><li>Develop a test Android application</li><li>Develop the Android library</li></ol><h3>Implementation</h3><h4>Data preparation</h4><p>First, let’s clarify what sensor and what kind of data can describe our motion gestures. Looks like, to precisely describe these gestures, both accelerometer and gyroscope should be used.</p><p>The accelerometer sensor obviously measures acceleration and, subsequently, movement:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*0WuCRHGqAh5FIiRa.png" /></figure><blockquote>The accelerometer has an interesting nuance; it measures not only the acceleration of the device itself but also the gravity of the Earth, which is roughly 9.8 m/s2. This means that the magnitude of the acceleration vector of the phone laying on the table will be equal to 9.8. Such values can’t be used directly and should be extracted from the Earth gravity value. This is not so easy a task, because it requires fusion of magnetometer and accelerometer sensors values. Hopefully, Android already has special Linear Accelerometer sensor that performs such calculations and returns corrected values.</blockquote><p>The gyroscope, on the other hand, measures rotation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*95fCez9-2tDDJm-S.jpg" /></figure><p>Let’s try to identify what values will correlate with our gestures. Obviously, in the accelerometer (meaning the linear accelerometer), X and Y values will highly describe the gestures. The accelerometer’s Z value is unlikely to be affected by our gestures.</p><p>As for the gyroscope sensor, it seems that only the Z axis is slightly affected by the gestures. However, to simplify implementation, I propose not including that value. In this case, our gesture detector will recognize moving the phone not only in the hand but also along a horizontal line — on the table, for example. This is presumably not too big a problem.</p><p>So, we need to develop an Android application that is able to record accelerometer data.</p><p>I developed such an <a href="https://github.com/ryanchyshyn/motion_gestures_detection/tree/master/GesturesManager">application</a>. Here is screenshot of the recorded “move right” gesture:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*MzziVNNcn3BTt0KI.jpg" /></figure><p>As you see, X and Y axes react very heavily to the gesture. The Z axis also reacts, but, as we stated, it will be not included in processing.</p><p>Here is the “move left” gesture:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*MwUx-aEc9pZOa_dm.jpg" /></figure><p>Note that the X values are almost opposite to the values from the previous gesture.</p><p>One more thing that should be mentioned is the data sampling rate. This reflects how often the data are sampled and directly influences the amount of data per time interval.</p><p>Another thing to consider is gesture duration. This value, like many others here, should be chosen empirically. I established that gesture duration is not longer than 1 second, but, to make things run smoother, I rounded it up to 1.28 seconds.</p><p>The data sampling rate I selected is 128 points per selected duration, which gives a 10-millisecond delay (1.28/128). This values should be passed into the <a href="https://developer.android.com/reference/android/hardware/SensorManager.html#registerListener(android.hardware.SensorEventListener,%20android.hardware.Sensor,%20int)">registerListener</a> method.</p><p>The idea, then, is to train a neural network to recognize such signals within a real-time data stream from accelerometer sensors.</p><p>So, next, we need to record a bunch of gestures and export to files. Of course, the same type of gestures (<strong>right</strong> or <strong>left</strong>) should be labeled with the same labels. It is difficult to say in advance how many samples are needed to train the network, but this can be determined by training results.</p><p>By clicking on the data, the sample duration will be automatically highlighted:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*xtAt63GpXeMDDyBO.jpg" /></figure><p>Now, the “save” button becomes enabled. Clicking it will automatically save the selection to the file in the working directory. The filename is generated in the form “{label}_{timestamp}.log.” The working directory can be selected using the application menu.</p><p>Also note that, after saving the current selection, the next gesture will be selected automatically. The next gesture is selected using a very simple algorithm: find the first X entry whose absolute value is greater than 3, then rewind backward 20 samples.</p><p>Such automation allows us to quickly save a bunch of samples. I recorded 500 samples per gesture. Saved data should be copied to a PC for further processing. (Processing and training directly on the phone looks promising, but TensorFlow for Android currently does not support training.)</p><p>In a screenshot provided earlier, the data range is roughly ±6. However, if you wave the phone more forcefully, it can reach ±10. It is better to normalize the data so the range will be ±1, which much better fits the neural network data format. So, I simply divided all data by a constant coefficient — in my case, 9.</p><p>The next step that can be performed before starting training is to filter the data to eliminate high-frequency oscillations. Such oscillations are not relevant to our gestures.</p><p>There are many ways to filter the data. One is to filter based on the <a href="https://en.wikipedia.org/wiki/Moving_average">moving average</a> box. Here is a picture of how it works:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*9SDbleL_Cwq4qfUK.jpg" /></figure><p>Note that X data maximum values are now half of the original. Since we will perform the same filtering of real-time data during recognition, this should be not a problem.</p><p>The final step to improve the training is data augmentation. This process expands the original dataset by performing some manipulations. In our case, I simply shifted the data to the left and right by several points:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*yLxpH1feArsCNVuV.jpg" /></figure><h4>Designing a neural network</h4><p>Designing a neural network is not a straightforward task and requires some experience and intuition. From the other side, neural networks are well known for some tasks, and you can simply adjust the existing network. Our task is very similar to the image classification task; the input can be treated as an image with a height of 1 pixel (and this is true — the first operation is reshaping that converts input of two-dimensional data [128 columns x 2 channels] to three-dimensional data [1 rows x 128 columns x 2 channels]).</p><p>So, the neural network input is [128, 2].</p><p>The neural network output is a vector with the length equal to the number of labels. In our case, it is [2]. The data type is double precision floating point.</p><p>Here is a schematic of the neural network:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*dHVW8-XrJjLkQbRn.jpg" /></figure><p>And here is a detailed schematic obtained through TensorBoard:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*zdxVUlEax9YiC4Qe.jpg" /><figcaption><a href="http://bit.ly/2BxNHCP">http://bit.ly/2BxNHCP</a></figcaption></figure><p>This schematic contains some auxiliary nodes needed only for training. Later, I will provide a clean, optimized picture.</p><h4>Training</h4><p>Training will be performed on a PC in a <a href="http://jupyter.org/">Jupyter Notebook</a> environment using Python and the TensorFlow library. Notebook can be launched in the <a href="http://conda.io/docs/user-guide/install/download.html">Conda</a> environment using the following configuration <a href="https://github.com/ryanchyshyn/motion_gestures_detection/blob/master/Python/environment.yml">file</a>. Here are some training hyperparameters:</p><p><em>Optimizer: Adam<br>Number of training epochs: 3<br>Learning rate: 0.0001</em></p><p>The dataset is splitted into the training and validation set as 7/3.</p><p>The training quality can be controlled by training and testing accuracy values. Training accuracy should approach but not reach 1. Too low a value will indicate poor and inaccurate recognition, and too high a value will lead to overfitting the model and may introduce some artifacts during recognition like non-zero recognition evaluation for non-gesture data. Good testing accuracy is a proof that a trained model can recognize unseen earlier data.</p><p>Training log:</p><pre>(‘Epoch: ‘, 0, ‘ Training Loss: ‘, 0.054878365, ‘ Training Accuracy: ‘, 0.99829739)<br>(‘Epoch: ‘, 1, ‘ Training Loss: ‘, 0.0045060506, ‘ Training Accuracy: ‘, 0.99971622)<br>(‘Epoch: ‘, 2, ‘ Training Loss: ‘, 0.00088313385, ‘ Training Accuracy: ‘, 0.99981081)<br>(‘Testing Accuracy:’, 0.99954832)</pre><p>TensorFlow graph and associated data are saved into files using the following methods:</p><pre>saver = tf.train.Saver()<br>with tf.Session() as session:<br>     session.run(tf.global_variables_initializer())<br>     # save the graph<br>     <strong>tf.train.write_graph(session.graph_def, &#39;.&#39;, &#39;session.pb&#39;, False)</strong></pre><pre>     for epoch in range(training_epochs):<br>     # train</pre><pre><strong>     saver.save(session, &#39;./session.ckpt&#39;)</strong></pre><p>Complete notebook code can be found <a href="https://github.com/ryanchyshyn/motion_gestures_detection/blob/master/Python/training.ipynb">here</a>.</p><h4>Exporting neural network</h4><p>How to save TensorFlow data was shown in a previous section. The graph is saved into a “session.pb” file, and training data (weights, biases, etc.) are saved into a batch of “session.ckpt” files. These files can be large enough:</p><pre>session.ckpt.data-00000-of-00001                 3385232<br>session.ckpt.index                               895<br>session.ckpt.meta                                65920<br>session.pb                                       47732</pre><p>The graph and training data can be frozen and converted into a single file suitable for running on a mobile.</p><p>To freeze it, copy the tensorflow/python/tools/freeze_graph.py file into your notebook directory, and run the following command:</p><pre>python freeze_graph.py --input_graph=session.pb \<br>     --input_binary=True \<br>     --input_checkpoint=session.ckpt \<br>     --output_graph=frozen.pb \<br>     --output_node_names=labels_output</pre><p>The resulting file is smaller than previous ones but still large enough:</p><pre>frozen.pb                                           1130835</pre><p>Here is this model in TensorBoard:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*7cNvQbrcwyZjMCRujJj2vQ.jpeg" /><figcaption><a href="http://bit.ly/2ppvaqo">http://bit.ly/2ppvaqo</a></figcaption></figure><p>To get this image, copy the <em>tensorflow/python/tools/import_pb_to_tensorboard.py</em> file into notebook directory and launch:</p><pre>python import_pb_to_tensorboard.py --model_dir=frozen.pb --log_dir=tmp</pre><p>where <strong>frozen.pb</strong> is a model file.</p><p>Now, start TensorBoard:</p><pre>tensorboard --logdir=tmp</pre><p>There are several ways to <a href="https://www.tensorflow.org/versions/master/mobile/prepare_models">optimize the model</a> for a mobile environment. To run described commands, you will need to compile TensorFlow from <a href="https://www.tensorflow.org/install/install_sources">sources</a>:</p><p>1. Removing unused nodes and general optimization. Execute:</p><pre>bazel build tensorflow/tools/graph_transforms:transform_graph<br>bazel-bin/tensorflow/tools/graph_transforms/transform_graph --in_graph=mydata/frozen.pb --out_graph=mydata/frozen_optimized.pb --inputs=&#39;x_input&#39; --outputs=&#39;labels_output&#39; --transforms=&#39;strip_unused_nodes(type=float, shape=&quot;128,2&quot;) remove_nodes(op=Identity, op=CheckNumerics) round_weights(num_steps=256) fold_constants(ignore_errors=true) fold_batch_norms fold_old_batch_norms&#39;</pre><p>Here is the TensorBoard result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*MsJj4mn-Em7UnOOz.jpg" /><figcaption><a href="http://bit.ly/2l805SI">http://bit.ly/2l805SI</a></figcaption></figure><p>2. Perform <a href="https://www.tensorflow.org/versions/master/mobile/optimizing">quantization</a> (convert floating point data format info 8 bit data format). Execute:</p><pre>bazel-bin/tensorflow/tools/graph_transforms/transform_graph --in_graph=mydata/frozen_optimized.pb --out_graph=mydata/frozen_optimized_quant.pb --inputs=&#39;x_input&#39; --outputs=&#39;labels_output&#39; --transforms=&#39;quantize_weights strip_unused_nodes&#39;</pre><p>As a result, the output file is 287129 bytes in size compared to the 3.5 Mb of the original files. This file can be used in TensorFlow for Android.</p><h4>Demo Android application</h4><p>To perform signal recognition in Android application you need to use TensorFlow library for Android. Add the library to gradle dependencies:</p><pre>dependencies {<br>implementation &#39;org.tensorflow:tensorflow-android:1.4.0&#39;<br>}</pre><p>Now, you can access the TensorFlow API through the TensorFlowInferenceInterface class. First, put the “frozen_optimized_quant.pb” file into the “assets” directory of your application (i.e., “app/src/main/assets”), and load it in the code (for example, on Activity start; however, as usual, it is better to perform any IO-related operations in background threads):</p><pre>inferenceInterface = new TensorFlowInferenceInterface(getAssets(), “file:///android_asset/frozen_optimized_quant.pb”);</pre><p>Note how the model file is specified.</p><p>Finally, look at how to perform the recognition:</p><pre>float[] data = new float[128 * 2];<br>String[] labels = new String[]{&quot;Right&quot;, &quot;Left&quot;};<br>float[] outputScores = new float[labels.length];</pre><pre>// populate data array with accelerometer data</pre><pre>inferenceInterface.feed(&quot;x_input&quot;, data, new long[] {1, 128, 2});<br>inferenceInterface.run(new String[]{“labels_output”});<br>inferenceInterface.fetch(&quot;labels_output&quot;, outputScores);</pre><p>Data is an input for our “black box.” There should be a flat array of accelerometer X and Y measurements, i.e., the data format is [x1, y1, x2, y2, x3, y3, …, x128, y128].</p><p>On the output, we have two float values with the range 0…1 with the probability of the input data according to “Left” or “Right” gesture. Note that the sum of these values is 1. Thus, for example, if the input signal does not match either Left or Right gesture, then the output will be close to [0.5, 0.5]. For simplification, it is better to convert these values into absolute 0…1 values using simple math.</p><p>Also, don’t forget to perform data filtering and normalization before running the recognition.</p><p>Here is final testing screen of the Demo application:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*NDbXbsO-u0WxO7eV.jpg" /></figure><p>where “red” and “green” lines are a real-time preprocessed signal. Yellow and cyan lines belong to “fixed” “right” and “left” probabilities, respectively. “Time” is processing time, and it is pretty low, which makes it possible to perform recognition in real time. (Two milliseconds means that processing can be run at a 500 Hz rate while we requested the accelerometer to update at 100 Hz).</p><p>As you can see, there are some surprising nuances. First, there are some non-zero probabilities even for the “silent” signal. Second, every gesture has long “true” recognition at the center with the value close to 1.0 and small opposite recognition at the edges.</p><p>Looks like, to perform accurate actual gesture recognition, some additional processing is needed.</p><h4>Android Library</h4><p>I implemented TensorFlow recognition altogether with additional processing of the output signal in a separate Android library. <a href="https://github.com/ryanchyshyn/motion_gestures_detection/tree/master/MotionGesturesDemo">Here</a> is the library and demo application.</p><p>To use it in your own application, add library dependency to the module gradle file:</p><pre>repositories { maven { url &quot;https://dl.bintray.com/rii/maven/&quot; } }</pre><pre>dependencies {<br>...<br>implementation &#39;uk.co.lemberg:motiondetectionlib:1.0.0&#39;<br>}</pre><p>create a MotionDetector listener:</p><pre>private final MotionDetector.Listener gestureListener = new MotionDetector.Listener() {<br>     @Override<br>     public void onGestureRecognized(MotionDetector.GestureType gestureType) {<br>        Log.d(TAG, &quot;Gesture detected: &quot; + gestureType);<br>     }<br>};</pre><p>and enable motion detection:</p><pre>MotionDetector motionDetector = new MotionDetector(context, gestureListener);<br>motionDetector.start();</pre><h3>Conclusion</h3><p>We passed through all steps of implementing motion gesture recognition on an Android application using the TensorFlow library: capturing and preprocessing training data, designing and training a neural network, and developing a test application and ready-to-use Android library. The described approach can be used for any other recognition/classification tasks. The resulting library can be integrated into any other Android application to boost it with motion gestures.</p><p>Hope you have found this article useful, you can also watch video overview below. <br>If you have a project idea in mind but don’t know where to start, <a href="https://lemberg.co.uk/contact-us">we’re always here to help you</a>.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FOn-plyyFR48%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DOn-plyyFR48&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FOn-plyyFR48%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f18529a1eefa37d02015a4068766564a/href">https://medium.com/media/f18529a1eefa37d02015a4068766564a/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9670249b8e9e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Agile Project Management through the looking glass]]></title>
            <link>https://medium.com/@WeAreLemberg/agile-project-management-through-the-looking-glass-a4d294c7113?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4d294c7113</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[drupal]]></category>
            <category><![CDATA[project-management]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[development]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Thu, 12 Oct 2017 13:02:22 GMT</pubDate>
            <atom:updated>2017-10-12T13:02:22.806Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Originally published at </em><a href="http://blog.lemberg.co.uk/agile-project-management-through-looking-glass"><strong><em>blog.lemberg.co.uk</em></strong></a><strong><em>.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*4yk1Z4K8zks6E-7dgr10vw.jpeg" /></figure><p>In IT industry, as in any other business segment, a lot of work related to the creation of a new product is being left behind the scenes. So let us shed light on some of it — project management.</p><p>I believe everyone has an idea who is a project manager: you see these guys with red eyes and nervous moves every day in the office. We are something like live communicators between the team and the client. We are those who fight for total happiness: so that everyone would not just work and check results, but also share part of the joy on the way. A Project manager is the person responsible for successful progress and delivery of the project (and also the one who is hated by all other involved parties, usually). But that is pretty general information and my idea for this article is to provide some details that might be helpful and interesting.</p><p>Over the past 3 years I used to use Agile methodologies for the projects I was responsible for, scrum in particular. So the following information is based on this experience.</p><h3>Sprint 0</h3><p>Work on the project starts long before the actual development begins. When sales managers are done with their part and all necessary legal issues are solved, the phase which we call “Sprint 0” is launched. If we set the reference to a classic project management model and 5 stages described in PM book, the two first stages — Conception &amp; Initiation and Definitions &amp; Planning — are referring to the activities that are supposed to be covered during Sprint 0. The “Todo” list for this described phase may be shortened or extended depending on the peculiarities of a certain project, here you can find an example, which is basically a compilation from several projects:</p><ul><li>Analysis of client’s requirements</li><li>Defining of the methodology that will be used</li><li>Resource planning, determination of the timeline and communication plan</li><li>Analysis of the market, for the purpose of creating a portrait of an end user</li><li>Writing user stories, forming the backlog</li><li>Prototyping and UI creation, feature design</li><li>Refinement of user stories and breakdown into development tasks</li></ul><h3>How does the magic happen?</h3><p>It should be noted that 2 activities from the list above are connected with a moment when the development team joins the game. These guys usually give input for designers, so that the concept isn’t just beautiful, but also compatible with a technical part and not too complicated in terms of implementation. But let’s get closer to the applied part: we usually have several teams that work on the project. In a way they compete a bit from time to time, but this results in a constructive discussion that helps to polish-up the product at each iteration, till the delivery point.</p><h3>Ready team</h3><p>These are professionals who are involved in the so called ‘Sprint zero’. They conduct analysis of the market, create user portraits, provide solutions and concepts, create architecture, think of feature design, work on prototyping, create the UI. Also work related to the gathering of primary acceptance criteria, as well as the writing of user stories, is done by them.</p><p>Here is one nice practice, which we use for feature design — the Slice &amp; Dice session. It is a meeting of architects where they “slice” a huge part of functionality into smaller features (breaking epics into user stories, if to use the Jira language) and “serve” those in the form of a task with a pretty clear description.</p><h3>Done team</h3><p>These guys work like restless ants on the actual development, in other words — code writing. They do the analysis part though: they go deeper into details in order to define the exact implementation steps and to write down how to test instructions for QA engineers.</p><p>Also they perform code reviews to make sure that everything written is in accordance to high standards; as well, they write automated tests, which help to keep the structure solid and can be ran further during testing process.</p><p>The Development process is divided into sprints, each from 2 to 4 weeks in length. Here is the list of activities, which we perform during each sprint:</p><ul><li>Sprint planning (where the development team and the product owner sit together to define the sprint goal and to decide which part of the project scope we should work on; also the acceptance criteria, implementation steps and estimations for each task are reviewed here)</li><li>Development</li><li>Refinement sessions (breakdown of user stories into smaller tasks, primary estimation)</li><li>Internal testing (provided by in-team QA engineers)</li><li>User acceptance testing (a check made on the client’s end)</li><li>Sprint review (meeting performed to check whether we reached the goal for the sprint and to demonstrate the work results)</li><li>Sprint retrospective (internal meeting with the team, performed with the purpose of monitoring the flow of the sprint and understanding what can be improved in future sprints)</li></ul><p>The team of guardians who don’t let the bug pass undiscovered. They run automated tests, proceed with manual tests (like pressing all reachable buttons both with fingers and noses at the same time). Also they perform cross-browser and cross platform testing for corresponding types of project. And of course they do accessibility tests to make sure that people with disabilities can also use the product.</p><h3>UAT team</h3><p>User acceptance testing is significant and necessary since on this stage we can make sure that the product is good (in terms of code quality), looks nice, etc. Yet, only feedback from end users will definitely show if it is desirable and wanted.</p><p>In order not to miss any necessary steps, we wrote the definition of done — the list of requirements for each user story, work on which can’t be completed before all these stages are passed. Here is an example of the definition of done:</p><ul><li>Code Complete and pull request passed the build</li><li>Code reviewed</li><li>Internal testing passed</li><li>Automated tests written and executed</li><li>Accessibility checked</li><li>UAT passed</li><li>Code Documentation in How To Test</li></ul><p>Once every task from the backlog has successfully passed the above mentioned list, we are ready to show the final version of the product to the world. It is time for the release (and keep in mind a good suggestion: never ever set it for Friday). Sure thing, that’s not the end: there will be further support and maintenance, as there is no finish line when it comes to perfection.</p><p>Now, I hope, you know a bit more about project managers, the process of product creation and some useful tips that I was happy to share here with you. However, you can still discover much more once you’ve become a part of this magic.</p><p>Hope you have found this article useful, don’t forget to leave your comments below the article. If you have a project idea in mind but don’t know where to start, <a href="https://lemberg.co.uk/contact-us">we’re always here to help you</a>.</p><p><em>Originally published at </em><a href="http://blog.lemberg.co.uk/agile-project-management-through-looking-glass"><em>blog.lemberg.co.uk</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4d294c7113" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HorseAnalytics. Dressage. iBeacons Implementation. Part 4]]></title>
            <link>https://medium.com/@WeAreLemberg/horseanalytics-dressage-ibeacons-implementation-part-4-5fc51088ddf4?source=rss-7925f49ef7c0------2</link>
            <guid isPermaLink="false">https://medium.com/p/5fc51088ddf4</guid>
            <category><![CDATA[android]]></category>
            <category><![CDATA[horses]]></category>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[beacons]]></category>
            <category><![CDATA[apps]]></category>
            <dc:creator><![CDATA[Lemberg Solutions]]></dc:creator>
            <pubDate>Wed, 11 Oct 2017 12:15:01 GMT</pubDate>
            <atom:updated>2017-10-11T12:15:39.966Z</atom:updated>
            <content:encoded><![CDATA[<p><strong><em>Originally published at </em></strong><a href="http://blog.lemberg.co.uk/horseanalytics-dressage-ibeacons-implementation-part-4"><strong><em>blog.lemberg.co.uk</em></strong></a><strong><em>.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/828/1*v4I5ne-JI1_r06iWFiGwww.png" /></figure><p>Theory: In terms of trilateration, user position is defined by accurate distance from the user to 3 predefined points. In general, trilateration is a three dimensional task and the user position is defined as a point of interception of three spheres.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*F5MnhTUx_dCqTbcV.png" /></figure><p>But in context of our task, all beacons and the user are approximately at the same altitude, so we can treat spheres as simple circles and use a two-dimensional coordinate system. Unfortunately, beacons provide us with quite a low accuracy, so the circles often have large interception areas or do not have them at all. So in real world we actually have to deal with such cases:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*JxyMxE7RNqdjSsVX.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*W3lYoSELOXy5Wkbx.png" /></figure><p>In case of a) and b) the most probable user location is a geometrical centre (green dot) of the triangle (red) produced by internal circle intersections.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*VgiUr3gqvn5roFNZ.png" /></figure><p>In case of c) user is probably situated out of a rectangle, so the location definition accuracy is low</p><p>And d) doesn’t suggest any solution at all</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*MZONdSJ2gpfKjmVo.png" /></figure><h3>Error Correction</h3><p>In order to get a solution in any case (Note: formally, b) and d) cases do not have any solution). We need to correct the distances in such a way, where all circles have at least some intersections (and common points).</p><p>There are generally 3 error sources:</p><ul><li>Environmental issues</li><li>Dynamic obstacles (other people/horses)</li><li>Device hardware accuracy barrier</li></ul><h3>Environmental</h3><p>Can be corrected by linear range correction, based on the approximate distance to the beacon (since such an error is probably influencing all beacons proportionally).</p><h3>Dynamic</h3><p>We can do nothing about dynamic obstacles, but just define which of beacons are inflated and exclude them from the calculations (such an error isn’t permanent). Additionally the error margin can be decreased by placing the beacons higher. The main drawback of the suggested solution however, is the fact that we’ll have to deal with a 3-dimensional position definition algorithm.</p><h3>Accuracy</h3><p>This error usually does have a random structure so it can be corrected simply by calculating the average value for some period of time. The main drawback of such a correction is the “position update lag”, depending on the analyzed timespan. Additionally such an error can be decreased by incrementing the number of beacons analyzed.</p><h3>Practical experiments</h3><p>In order to perform tests we’ve made an app, which performs according to 4 stages:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/851/0*c7HGASsKDpp2Mmwy.png" /></figure><p>The first stage has 3 alternative approaches:</p><p>a) Linear accuracy correction (circles are extended proportionally to their current values). Targeted to resolve environmental error issue</p><p>b) Reliability-based correction (here we have implemented a complex reliability criteria for circle values, the algorithm was able to increase or decrease circle radiuses to get the user position region as small as possible.) The algorithm was targeted to resolve environmental and dynamic obstacle issues. Such behaviour allowed us to analyze all beacons available, and not just the nearest 3.</p><p>c) Reliability-based correction (simplified). Similar to the previous, but uses a simplified error correction norm definition, which allows for faster calculations and the ability to remove some critical points, yet doesn’t allow for circle-size decreasing. Therefore, it can’t correct decremental errors (such errors are produced by hardware anyway) from the algorithm. Estimated accuracy loss (compared to b)) is minimal.</p><p>Note: the current algorithm requires an update to handle circle positioning c) correctly. In all experiments, the beacons (A,B,C) were positioned to form an equidistant rectangular triangle. So that AB = AC. We’ll call it “ds” in future.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/0*SZwv39886efRy3uN.png" /></figure><h3>Experiment Logs</h3><p>Indoor Test #1</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*M6a7dhpshtcbF9d2.png" /></figure><p>Positioning returns unstable values. The main cause seems to be high beacon broadcast power (seems that short range tracking provides better accuracy. No low broadcasting power)</p><p>Indoor Test #2</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*RJBY7TApOKkmskQ7.png" /></figure><p>Positioning returns reliable data (accuracy about 0.5–1m) in the case when the user is situated on ABC triangle sides. Proximity to the beacons is detected accurately. Some randomness is present. Side spectators (dynamic obstacles) make the algorithm unstable. Position definition outside of the triangle (circle combination c) ) doesn’t work.</p><p>Indoor Test #3</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*NgH5H8TjOloE_LGa.png" /></figure><p>Accuracy is slightly better compared to test 2, but spectators increase the error significantly and the beacon is sometimes lost (user position is undefined due to error).</p><p>Indoor Test #4</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*x22AAHIPr1SnJ1AZ.png" /></figure><p>The algorithm can’t find the user position, random data is returned. Algorithm correction is required.</p><p>Indoor Test #5</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*wWu0pkUQIy92YPEK.png" /></figure><p>The algorithm can find the user position with an approximate accuracy of 1–1.5m. Proximity to the beacon is defined correctly. Accuracy and stability is lower compared to correction a).</p><p>Indoor Test #6</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*qbioTgmGedykIFPd.png" /></figure><p>Positioning is showing inaccurate data. User location is often lost. The algorithm requires an update to support more than 3 beacons.</p><p>Outdoor Test #1</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/0*VAyQCrwp7vtOBU4s.png" /></figure><p>The algorithm is showing 0.5–1.5m accuracy on ABC triangle sides and in close proximity to the beacon. The position outside of the triangle can’t be defined.</p><h3>Conclusion</h3><p>The algorithm might be usable (in context of section definition) in case we decide to track just one further section, which the user has to move into.</p><p>Currently, the amount of beacons which can be used, is limited to 3 beacons. However, there is a possibility to overcome this limitation by combining the 3 beacons, analyzing all possible subsets and then choosing the most optimal subset placement. The current implementation might trigger a “user section update” before he actually changes it. And large-scale field test is required.</p><p>Hope you have found this article useful, don’t forget to leave your comments below the article. If you have a project idea in mind but don’t know where to start, <a href="https://lemberg.co.uk/services/iot-development-services">we’re always here to help you</a>.</p><p><strong>SEE ALSO:</strong></p><ul><li><a href="http://blog.lemberg.co.uk/horseanalytics-dressage-more-accurate-positioning-task-part-1">HorseAnalytics. Dressage. More Accurate Positioning. Task. Part 1</a></li><li><a href="http://blog.lemberg.co.uk/horseanalytics-dressage-ibeacons-approach-part-2">HorseAnalytics. Dressage. iBeacons Approach. Part 2</a></li><li><a href="http://blog.lemberg.co.uk/horseanalytics-dressage-ibeacons-implementation-part-3">HorseAnalytics. Dressage. iBeacons Implementation. Part 3</a></li></ul><p><em>Originally published at </em><a href="http://blog.lemberg.co.uk/horseanalytics-dressage-ibeacons-implementation-part-4"><em>blog.lemberg.co.uk</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fc51088ddf4" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>