<?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 Vijay Betigiri on Medium]]></title>
        <description><![CDATA[Stories by Vijay Betigiri on Medium]]></description>
        <link>https://medium.com/@vijay.betigiri?source=rss-6139f9655848------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1YZQKyMenHB6hveeIjMowA.jpeg</url>
            <title>Stories by Vijay Betigiri on Medium</title>
            <link>https://medium.com/@vijay.betigiri?source=rss-6139f9655848------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 10:04:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vijay.betigiri/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[The Analogous Journey of Modernizing an Old Luxury Car and Legacy Application]]></title>
            <link>https://medium.com/@vijay.betigiri/the-analogous-journey-of-modernizing-an-old-luxury-car-and-legacy-application-97b81cb3abb3?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/97b81cb3abb3</guid>
            <category><![CDATA[analogy]]></category>
            <category><![CDATA[transformation]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[legacy-modernization]]></category>
            <category><![CDATA[application-modernization]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Mon, 24 Apr 2023 07:27:38 GMT</pubDate>
            <atom:updated>2023-04-24T11:48:10.567Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NdxlOeUJmj32U2pkNYZMYA.png" /></figure><p><strong>Introduction:</strong></p><p>We often use the analogy of an old car to describe legacy applications. The analogy is easy to understand and relatable even for a non-technical person. In this article, I have attempted to expand the analogy.</p><p>In today’s digital landscape, companies want to increase productivity, become agile and gain a competitive edge. One way to achieve this is by modernizing legacy applications to the cloud. However, why is this necessary? Let’s use the analogy of an old, classic car to understand the process better.</p><p><strong>The car analogy:</strong></p><p>Legacy applications are like old luxury cars. They were built using outdated technologies and requires constant maintenance, making them slow, unreliable, and expensive to maintain. Similarly, old cars require frequent oil changes, tune-ups, and repairs, which are often difficult to find in the market. Despite their outdatedness, legacy applications still have value, such as embedded data and knowledge, and provide core business functionality.</p><p>But like an old luxury car, legacy applications can’t meet the modern world’s expectations of being fast and agile. So what can you do? You cannot buy a new car or throw away the old one. Instead, you could take it to a garage, where a skilled mechanic will gradually upgrade the engine, replace worn-out parts, and add new safety features to make it faster, safer, and more efficient.</p><p>In application modernization, companies can take their legacy applications to the cloud. An experienced cloud architect will use the right tools and processes to modernize the applications. Patterns such as microservices, serverless computing, cloud-native PaaS, DevOps, and data modernization will be used to improve the application’s performance, scalability, and availability. We will extend the analogy to each of the patterns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1012/1*BOzPfs63Idpl4j8OeLT-5g.jpeg" /></figure><p><strong>Microservices is like a modular electric car:</strong></p><p>So let’s continue with our car analogy to understand micro services. Imagine you have an old car that has a complex, monolithic engine with many interdependent parts. If one part fails, it can affect the entire engine, and replacing or fixing the part can be time-consuming and costly. Similarly, legacy applications often have a monolithic architecture, where all the components are tightly coupled and interdependent. Any change or update to one component can affect the entire application, making it difficult to maintain and scale. An engine repair means the car cannot be driven for few days.</p><p>Now, let’s compare the old car with a modern electric car. An electric car has a modular design, where each component is designed to be independent and replaceable. If a battery cell fails, it can be easily replaced without affecting the rest of the battery pack. Some electronic components will be micro chips on a PCB. Similarly, microservices architecture is designed to be modular, with each service being designed to perform a specific function or task. If one microservice fails or needs updating, it can be easily replaced or updated without affecting the rest of the application.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1015/1*meXg17w4mNshHZg1BiRibw.jpeg" /></figure><p><strong>Serverless Computing is like a Smart Engine:</strong></p><p>Imagine you have a high-performance sports car with a powerful engine, but you don’t use it everyday. When you’re driving in the city or stuck in traffic, you don’t need all that power, and it’s wasteful to keep the engine running at full capacity all the time. Similarly, in traditional application architectures, servers are constantly running, even when they’re not being utilized to their full potential.</p><p>Now, imagine that you could install a smart engine in your sports car that automatically adjusts its power output based on your driving conditions. If you’re cruising on the highway, it uses less power, and if you need to accelerate quickly, it would provide more power. This is similar to serverless functions, which only run when they’re triggered by an event or request, and then automatically scale up or down to handle the workload.</p><p>In addition, you are gifted with a plan that you don’t need to manage the sports car maintenance which is a major worry for the user.</p><p><strong>Cloud native PaaS are the ready to use features of a modern car:</strong></p><p>Public cloud providers offer platform services so that we do not have to build our own services. Some examples of PaaS services are authentication (e.g. Azure Active Directory), Managed kubernetes, Monitoring services, DevOps etc.</p><ul><li>Let’s start with authentication. Authentication is like the lock and key system in a car. Just like a lock and key system allows you to secure your car and prevent unauthorized access, authentication enables you to secure your applications and protect sensitive data from unauthorized access. With cloud-native authentication services, you can implement secure authentication and authorization mechanisms easily, just like you can lock and unlock your car with the push of a button.</li><li>Kubernetes is like the smart engine management system in a car. The engine management system controls the performance, efficiency, and reliability of the engine by managing the fuel injection, ignition timing, and other critical parameters. Similarly, Kubernetes manages the performance, scalability, and reliability of containerized applications by automating deployment, scaling, and management tasks. Just like an engine management system enables a car to run smoothly and efficiently, Kubernetes enables applications to run smoothly and efficiently in a cloud-native environment.</li><li>Monitoring services are like the navigation system in a car. A navigation system provides real-time information about your surroundings, helps you navigate to your destination, and provides guidance along the way. Similarly, a monitoring service provides a real-time view of the terrain and alerts if something is not going well.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N-oJ0VY1THoIJJZNEPLEnA.jpeg" /></figure><p><strong>DevOps is like a pit crew in a car race:</strong></p><p>In a race, a pit crew is responsible for quickly servicing the car between laps to optimize its performance and keep it running smoothly.</p><p>Similarly, in software development, DevOps teams work to optimize the software development and deployment process to keep it running smoothly.</p><p>Just like a pit crew uses specialized tools and techniques to quickly and efficiently service the car, DevOps teams use specialized set of tools and processes to quickly and efficiently deploy software. They use automation tools to streamline the process of building, testing, and deploying code, just as a pit crew uses specialized tools to change tires, refuel the car, and make minor repairs.</p><p>DevOps teams also focus on continuous improvement in agile sprints, just like a pit crew analyzes data from each pit stop to make adjustments and improve their performance for the next lap. DevOps teams collect data on the software development and deployment process, analyze it, and make adjustments to improve performance, reliability, and efficiency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1015/1*4JAOf3Dw8IdnucoDJDP3ow.jpeg" /></figure><p><strong>Preserving the Intrinsic Value of Legacy Data while Modernizing:</strong></p><p>Just as the aesthetic, charm, and character of the classic car are invaluable to its owner and enthusiasts, legacy data also have unique value and insights that cannot be found elsewhere.</p><p>Modernizing an old car can enhance its performance, reliability, and safety while maintaining its classic charm. Similarly, modernizing legacy data can enhance its accessibility, reliability, and security while preserving data’s intrinsic value and significance.</p><p>With database freedom, it is possible to get rid of heavy licensing costs and move to open-source and managed databases on the public cloud.</p><p><strong>Conclusion</strong></p><p>While the car analogy may have its limitations, it can be a useful tool in helping someone to understand the concept of application modernization, especially for those who are not familiar with the technical jargon used in application modernization. By using relatable examples like the car analogy, we can help demystify the process of application modernization and make it more accessible to a wider audience.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=97b81cb3abb3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The G.O.D. Framework: A Balanced Approach to Building Autonomous AI Systems]]></title>
            <link>https://medium.com/@vijay.betigiri/the-g-o-d-framework-a-balanced-approach-to-building-autonomous-ai-systems-99ce6c7a60a3?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/99ce6c7a60a3</guid>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Wed, 05 Apr 2023 16:27:13 GMT</pubDate>
            <atom:updated>2023-04-06T09:13:09.451Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>Introduction:</strong></h4><p>Autonomous AI systems are intelligent computer programs that can autonomously perform tasks and make decisions without human intervention or guidance. The development of efficient autonomous AI systems presents challenges due to biases, ethical considerations, transparency, security, and control. In this article, we explore the theoretical G.O.D. (Generator, Operator, Destroyer) framework inspired by mythology as a way to build an autonomous AI system that revolutionizes our lives through a balanced approach to problem-solving and decision-making.</p><p>The G.O.D. framework echoes the Hindu Trimurti, consisting of Brahma, Vishnu, and Shiva — the Generator, Operator, and Destroyer forces, respectively. In the G.O.D. framework, three AI models have unique roles and responsibilities: the Generator AI model creates tasks based on a goal, the Operator AI model executes tasks autonomously and efficiently, and the Destroyer AI model monitors progress and takes corrective actions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xp2zT4Pg4e2uKmWI7OzeqQ.jpeg" /></figure><p><strong>The Workflow of the G.O.D. Framework in Autonomous AI Systems</strong></p><p>1. Define the goal: A human user defines the AI system’s goal.</p><p>2. Generate tasks: The Generator creates clear and unambiguous tasks aligned with the system’s goal.</p><p>3. Operator execute tasks: The Operator autonomously and efficiently executes tasks, potentially replicating itself to perform tasks in parallel.</p><p>4. Destroyer tasks: The Destroyer monitors task progress and takes corrective actions to refine/ purge tasks and ensure the system works towards the desired outcome.</p><p>5. Feedback communication and coordination: The Destroyer refines tasks and ensures the system works towards the desired outcome.</p><p><strong>Teacher tasks:</strong></p><p>Embodying the spirit of Lord Dattatreya, the Guru of all G.O.D.s in Hindu tradition, the Teacher model represents the importance of continuous learning and improvement. The Teacher model learns from experiences and adjusts the other three models’ roles accordingly. Additionally, it checks for patterns of unethical goals/ task creation and controls it.</p><p><strong>Implementing the G.O.D. framework</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OfttmAk72n9qHFg0ewNC8g.png" /></figure><p><strong>Generator AI model:</strong></p><p>The Generator AI model uses techniques such as natural language processing, machine learning, and deep learning to create specific tasks based on the human user’s defined goal. Leveraging recent developments in Large Language Models (LLMs) like GPT-4, the Generator model can be deployed as a serverless function on public cloud platforms, triggered when a goal message appears from the user or the Destroyer model requests task refinement/removal.</p><p><strong>Operator AI model:</strong></p><p>The Operator AI model executes tasks generated by the Generator AI model autonomously and efficiently, using techniques such as robotics, computer vision, and natural language generation. With access to vast internet data, the Operator model can also be deployed as a scalable serverless function, triggered by messages in the task queue. LLMs like GPT-4 can be utilized for tasks that involve text outputs or can be described in the text.</p><p><strong>Destroyer AI model:</strong></p><p>The Destroyer AI model monitors task progress and takes corrective actions to refine tasks and ensure the system works towards the desired outcome. It employs techniques such as reinforcement learning, anomaly detection, and decision-making algorithms for continuous improvement and refinement. Serverless functions and models like GPT-4 can be considered for deployment and processing to avoid building custom models.</p><p><strong>Teacher AI model:</strong></p><p>The Teacher model monitors the performance of the other three models (Generator, Operator, and Destroyer) and provides corrective actions to prevent repetitive loops and ensure ethical alignment with guidelines and goals. The Teacher model should be trained on a large amount of data covering a broad range of scenarios and use cases. Existing tools like OpenAI moderation endpoints can be useful for part of it, and can save on model training costs.</p><p><strong>Example goal:</strong></p><p>We will understand this with an example. We will use GPT models to deliver this example goal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lxvyn00ZPJDeWNZ-5a_FWw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e02FLN0KpZSLeO5DY-Rq-g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7jaIwvs0gXN9f6om8IcXYg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cqpOimf0ORmUJdRlkiHbMw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RvjNRLrLyS3d1RJG22EPgg.png" /></figure><p><strong>Benefits of the proposed approach</strong></p><p>1. Simplicity: The G.O.D. framework offers a straightforward and easy-to-implement approach for ethical AI, leveraging successful LLMs.</p><p>2. Balanced roles: The framework balances the four models (Generator, Operator, Destroyer, and Teacher), enabling the AI system to continuously learn and improve without entering repetitive loops or unethical behaviour.</p><p>3. Ethical decision-making: The G.O.D. framework emphasizes ethical decision-making and goal-setting, helping to prevent harmful or unethical behaviours.</p><p>4. Continuous learning: The presence of Lord Dattatreya and the Teacher model emphasize continuous learning and improvement, aiding the system’s adaptability to change circumstances.</p><p>5. Flexibility: The G.O.D. framework is designed for flexibility, with the Generator, Operator, and Destroyer models allowing for quick adjustments to goals and tasks as needed.</p><p>6. Improved performance: By emphasizing goal-setting and task optimization, the G.O.D. framework may enhance the overall performance of the autonomous system, leading to better user outcomes.</p><p><strong>Limitations</strong></p><p>1. Complexity: Implementing the G.O.D. framework with custom models requires expertise in AI, ethics, and other domains.</p><p>2. Computational Requirements: The G.O.D. framework involves multiple AI models, potentially requiring significant computational resources, leading to high costs and slower processing times.</p><p>3. Ethical Bias: Despite the G.O.D. framework’s aim to prevent unethical behaviour, AI models may exhibit ethical bias due to data biases or subjective interpretations of ethical values.</p><p>4. Limited Generalization: The G.O.D. framework may work well in specific domains but may not generalize well to other domains, limiting its broader applicability.</p><p>5. Limited Flexibility: The G.O.D. framework may not be flexible enough to adapt to changing situations or unexpected events, potentially limiting its ability to make appropriate decisions in complex scenarios.</p><p><strong>Conclusion</strong></p><p>The G.O.D. framework provides a simple yet effective approach to implementing ethical AI systems. Incorporating the four models of Generator, Operator, Destroyer, and Teacher, the system is designed to balance between achieving goals and avoiding unethical behaviour. The presence of the teacher model represents the importance of continuous learning and improvement in the system. While there are limitations and challenges to implementing any autonomous AI system, the G.O.D. framework offers comparative benefits over similar systems and is a step towards building more responsible and ethical AI systems.</p><p><strong>Disclaimer:</strong></p><p>The G.O.D. framework uses Hindu mythology deities as a symbol of different roles involved in a balanced system. However, it is important to note that the framework is not intended to promote or hurt the sentiments of any specific religion or belief system. It is simply a framework proposed to address the various issues faced with the development and deployment of AI systems.</p><p><strong>References</strong></p><ol><li>Hindu mythology: Various Puranas- Brahma Purana, Vishnu Purana, Shiva Purana, Avadhut Gita</li><li><a href="https://github.com/yoheinakajima/babyagi:">https://github.com/yoheinakajima/babyagi:</a> This framework developed by <a href="https://twitter.com/yoheinakajima">@yoheinakajima</a> also proposes creation, prioritisation, and execution agents with memory.</li><li>European Commission. (2021). White Paper on Artificial Intelligence: a European Approach to excellence and Trust. Retrieved from <a href="https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en">https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en</a>.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=99ce6c7a60a3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Gardener Chatbot]]></title>
            <link>https://medium.com/@vijay.betigiri/ai-gardener-chatbot-8e5f2d3d3d1c?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e5f2d3d3d1c</guid>
            <category><![CDATA[gardening]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[farmers]]></category>
            <category><![CDATA[amazon-lex]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Sun, 14 Jan 2018 15:56:32 GMT</pubDate>
            <atom:updated>2018-01-15T06:39:50.998Z</atom:updated>
            <content:encoded><![CDATA[<p>It is not easy for any gardener to see their garden plants suddenly losing their vigor. But trying to figure out the problem behind it is another matter. Sometimes it’s obvious — the snail’s still there munching away, but most of the times the gardener will be left with only a few clues like discoloration of leaves or some unusual white or brown dots. The problem can be anything — air-borne pest, nutrition, under/ over-watering, soil condition, soil-borne disease and many more. Getting a timely and professional help for gardening is not easy. Because of this, many gardeners lose their plants and eventually their interest in gardening.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/970/0*K2-7ZVofeDpmTXVV.jpg" /></figure><p>We (me with my friend Vishwas Salunkhe) developed a fun project to explore the chatbot functionality. The gardener chatbot can help with following,</p><ul><li>Recognize the problems by visual inspection such as pests, nutrition deficiency</li><li>Advice with fertilizers, pesticides</li><li>Receive product orders</li><li>Set action reminders like spray pesticide, watering</li></ul><p>Here is a short video demo of the chatbot.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FgVhnI4ecmlU%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DgVhnI4ecmlU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FgVhnI4ecmlU%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/8cc62bef9c05f017359ab306758bd226/href">https://medium.com/media/8cc62bef9c05f017359ab306758bd226/href</a></iframe><p>Building such a solution is not as difficult as it seems. Let us understand why what and how of it,</p><h4>What and Why of Chatbot</h4><p>A chatbot is an application designed to simulate human conversation over a chat messenger like facebook, slack. Chatbot promises to be a replacement for most of the traditional B2C applications. Some of the problems with existing mobile apps are,</p><ul><li>The user needs to download and install the application. It consumes precious storage space of the mobile.</li><li>The developer finds it difficult to reach their target customers because of too may apps in the marketplace. There are already 30 million+ apps in the market and growing!</li><li>Almost half of the users stop using the app during their first usage. 77% users uninstall the app within first 3 days and 95% of the users stop using the app in 3 months of installation.</li><li>The cost of building an app is high, about $270,000 on an average.</li></ul><p>Chatbot can replace the apps because of their following features,</p><ul><li><strong>Easier to reach customers:</strong> Developers can leverage on billions of messenger users.</li><li><strong>No Installation: </strong>No separate app installation is required. The user can use messenger over mobile or web. In case of facebook chatbot, the user just needs to find the facebook page and start chatting.</li><li><strong>Always ready:</strong> The chatbot can be live 24 x 7 x 365 and provide service to any number of users without getting tired.</li><li><strong>Easy to use: </strong>User doesn’t need to learn the UI or app. They can communicate with the bot by uttering their sentence in natural language and get the response.</li><li><strong>Easy to build:</strong> The developer doesn’t need to build UI. No need to learn new coding languages, in fact, most of the functionality can be built without any coding. The developers don’t have to worry about scalability, availability etc.</li><li><strong>Multi-lingual: </strong>Developer can integrate the chatbot with google (or other) translation service so that the chatbot can detect the language and respond in the same language. Google supports more than 100 languages. A developer doesn’t have to put any extra efforts for featuring multiple language<strong>s.</strong></li><li><strong>More engaging: </strong>By being a human-like friendly adviser, app stickiness (engagement and retention) increases tremendously.</li></ul><p>Surely the chatbot can revolutionize the app development and delivery. Gartner has predicted that by 2020, customers will manage 85% of their relationship with the enterprise without interacting with a human. A chatbot is going to play a major role in this relationship.</p><h4>How to build a chatbot</h4><p>There are many platforms for the developers to choose from. Google Dialogflow, Facebook wit.ai, Amazon Lex, IBM Watson, Chatfuel are some of the leading ones. Dialogflow is my favorite option but this chatbot was developed using Amazon Lex for trying it out. The development process would have been similar in any other platforms.</p><p>Chatbot platform provides following built-in functionality and provides easy ways for the developer to configure their application logic.</p><ul><li><strong>Messenger integration: </strong>A chatbot provides integration with messenger such as facebook, slack, Twilio (for SMS), or customized mobile app. For the gardener demo, we used Slack messenger because of its group chat feature. This can be useful for building a community helped by a gardener chatbot.</li><li><strong>NLP, the backbone:</strong> A Natural Language Processing (NLP) based conversation service forms the backbone of the chatbot. The NLP will be trained using a lot of conversations. The functionality of this conversation engine is to know the context of the discussion and extract an intent and entities from the user’s sentence.</li></ul><p>The user can utter their sentence in different ways,</p><blockquote><em>…..</em></blockquote><blockquote><em>Chatbot: You need to use 19:19:19 NPK fertilizer for nutrition.</em></blockquote><blockquote><em>User: OK, order 2 kg of it</em></blockquote><p>The chatbot understands the context is about fertilizer without the user explicitly mentioning it. Alternatively, the user may say,</p><blockquote><em>User: Can you please deliver 2 kg of 19:19:19 NPK fertilizer to my registered address. ….</em></blockquote><p>In this conversation, the chatbot extract intent as ‘Purchase Fertilizer’ and Entities as Product- 19:19:19 NPK fertilizer, Address- Registered address, Qty- 2 kg. In case the user has not mentioned any of the required entities, the chatbot can ask for these details.</p><p>In order to train the conversation engine, developer needs to provide few possible sample utterances for each intent. (e.g. For buy_a_product, the developer may provide utterances as ‘Buy it for me’ or ‘Buy neem oil for me’ or ‘I want this product’). The developer needs to separate out the entities (e.g. product name) in each utterances.</p><ul><li><strong>Backend functionality:</strong> In some cases, it is not sufficient to have a bi-directional dialog between the user and chatbot. The chatbot may also need to process the user’s request by implementing it’s own logic or calling external service. For example a request as “How is weather in Bangalore Today?” will need an API call to weather.com api by using entities as city=Bangalore, Date=Today. Following are the functionalities built for gardener bot.</li></ul><ol><li>Ordering fertilizer from e-commerce store by calling appropriate APIs.</li><li>Setting a reminder for watering, pesticide.</li><li>Symptom recognition (described below).</li><li>Providing advice on products, best practices</li></ol><ul><li><strong>More ways to engage: </strong>The chatbot can also be integrated with text-to-speech and speech-to-text converter, tone analyzer. This will allow voice conversation with the chatbot like Siri or google assistant.</li></ul><h4>Symptom Recognition</h4><p>We developed Convolutional Neural Network (CNN) to identify the issue by taking infected plant leaf photo as input and predicting the pest/ nutrition deficiency. We trained the CNN with five visually differentiable pest images (aphids, mealy bugs, scale, leaf miner) and three nutritional deficiency issues such as Nitrogen, Phosphate, Magnesium. We achieved about 82% accuracy with a limited dataset.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Q7Jwj9R2LPpAw_dp.jpg" /><figcaption>Mealybug infested leaf</figcaption></figure><h4>What’s more?</h4><p>Indian farmers find it difficult to get an instant professional advise for their daily farm related issues in their own language. Many times this results in losing their crops. With a wide smartphone spread in India, the advice can be just an utterance away. They can even order their products and get advice on the product usage.</p><p>If you have liked this article, please clap for me.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e5f2d3d3d1c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI based UI Development (AI-UI)]]></title>
            <link>https://medium.com/@vijay.betigiri/ai-based-ui-development-ai-ui-100deb28f54d?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/100deb28f54d</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[jobs]]></category>
            <category><![CDATA[ui]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Sun, 07 Jan 2018 07:20:10 GMT</pubDate>
            <atom:updated>2018-01-08T05:28:09.883Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Artificial Intelligence (</em><strong><em>AI</em></strong><em>) is currently one of the most popular topics in the industry with seemingly endless applications in everything from matchmaking to self-driving cars. The most disturbing aspect of AI that we hear is it will result in massive job losses across industries. Can AI also affect the IT jobs? If so which skills will be impacted? When? How? These are some questions every software engineer must be seeking.</em></p><p>Creative designers or business users comes up with UI (User Interface) ideas for application/ website on a sheet of paper or on a whiteboard or on their fancy graphics tablet. It is a job of an UI developer to convert the design idea/ wireframes into a working UI keeping the creative design intent in mind. This is one of the complex, time-consuming step in software development process. In this article, we will see an interesting example of applying AI for UI development. We will try to understand this by comparing it with human learning process and (over)simplifying the technology behind it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*1uX-Dj89sDWnZyaS.jpg" /><figcaption>Typical hand drawn design for a UI</figcaption></figure><h4>Mimicking our eyes and brain</h4><p>As a child, we learn to observe and label the things around us. The learning happens through feedback provided by our parents and others. Our brain gets trained to look for some pattern, texture, color, size in a object to identify it. In AI, Convolutional Neural Network (CNN) is a class of deep neural network very effective at recognizing the objects in a given image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/864/1*t0paWY9V-OjpXQQ41pmuQQ.png" /></figure><p>Basic idea behind CNN is to look for some shapes or patterns with the help of various filters in small parts of the image one at a time. Below figure shows applying 2 filters to look for slanted lines. Based on the filter results features are extracted. Finally by voting for the extracted features, the algorithm can conclude on the objects in the image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*Y0tvNnH7sXHw2zP7TrCKgA.gif" /></figure><h4><strong>Describing the image</strong></h4><p>The child starts uttering a single word label for each identified object, such as ‘ball’. Soon she will also learn to identify the relationship between the identified objects and describe it in a short sentences such as ‘a red ball and a brown bat is on the lawn’. The learning happens through a cycle of trial and errors.</p><p>In AI, for a given image constructing sentences from the word labels is a job of LSTM (Long Short Term Memory) networks. This process is called as image captioning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/1*xPoSq0VS7TwpYGA39j_swg.png" /></figure><p>Below are some examples of AI based image captioning. More such examples are at <a href="http://cs.stanford.edu/people/karpathy/deepimagesent/">http://cs.stanford.edu/people/karpathy/deepimagesent/</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*6BFOIdSHlk24Z3DFEakvnQ.png" /></figure><p>The image captioning is achieved by appending LSTM network to the CNN discussed earlier. LSTM is very effective in language related tasks, because of their unique property of referring to their previous outputs. LSTM generates a word at a time. The next word is decided based on it’s inputs, but also on previous words generated. e.g. in a sentence ‘My name is John.’, you can say ‘John’ only if earlier three words were ‘My name is’. The sequence of words forms into a sentence. Like any other neural network, LSTM goes through a learning at building the sentences.</p><h4>UI Development Process</h4><p>Typically UI development happens through following steps,</p><ul><li>Creative designers or business users of the application likes to hand draw their UI design ideas on a whiteboard or a graphic tablet or even a piece of tissue paper.</li><li>Designer uses wireframing tool on a computer to create the same design again. This is a redundant step.</li><li>UI developers will translate the wireframes into a working UI code. The developers and designers goes through a iterative process till the expected UI is built. This step is a time consuming and repetitive process.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/861/1*ECgxalZYXidW6HjorPxDig.png" /></figure><h4>AI based UI development</h4><p>What if the hand-drawn design idea is directly translated to a working UI? AI can do this. Below is an example of the same.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/1*Vrv1PInZtKegTkMAC-HY2w.png" /><figcaption>UI generated with pix2code</figcaption></figure><p>In image captioning, AI describes objects (such as dog, horse) in a scene and builds a English sentence describing the objects and their relationship with each other.</p><p>In case of UI code, the UI design is like a scene, but instead of dog and horse will have UI objects like button, slider. Instead of English language, the objects will be described in UI code. The UI code is having a limited vocabulary (such as button, slider), and relationship between objects are described with few more words (such as position, hierarchy). Thus UI code generation can be considered specific use case of image captioning.</p><p>UI code generation goes through two stages.</p><h4>Training Stage:</h4><p>Imagine a child (child_1) learning to look at many UI images and creating a list of the UI objects for each UI image. Other child (child_2) learns to read the descriptive code for the same UI. Third child (child_3) learns to find the relationship between the child_1 and child_2’s learning. They together learn to observe a image and create a corresponding UI code.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/393/1*_UWquzIdb6orS7pNcfTZUg.png" /></figure><p>CNN takes role of Child_1, LSTM as Child_2 and another LSTM as Child_3. <em>(For a complete technical explanation, refer the link for pix2code paper at end of the article.)</em></p><h4>Sampling Stage:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/405/1*09gwoHvgPR3IIg3WrWoqnw.png" /></figure><p>The trained model is now ready to process hand drawn GUI drawing. The code context is updated for each prediction to contain the last predicted token. The resulting sequence of DSL tokens is compiled to the desired target language (e.g. for android, iOS, HTML etc.) using traditional compiler techniques.</p><h4>Benefits of AI-UI</h4><ul><li>For designers and developers, AI based solution would save critical time early on a project by rapid prototyping, boost iteration cycles, and eventually enable the development of better apps.</li><li>They will save on all the trivial, repetitive and redundant tasks.</li><li>It also will allow designers and developers to focus on what matters the most that is to bring value to the end-users.</li><li>The entry barrier to build apps will become really low. Learning to use a UI design tool takes time, learning to code takes even more time. However everyone can draw UI on paper. This will allow your grandma to go from an idea to a working UI running on her phone in a matter of seconds.</li></ul><h4>Current and future state</h4><p>As of now only few AI based UI development products (e.g. Uizard) are getting developed and not yet reached maturity to replace the human UI developers. But still they are good as an assistant for any UI developers. In coming years, we may see new approaches and improved AI products, where this assistant will take over the role of the experienced UI developer. It’s time for UI developers to look at the changing trends and get ready for Reskilling.</p><p>Still many of us may think generating UI code from the creative designers drawings is OK, but AI itself cannot come up with it’s own creative UI designs. We still need artists, creative designers, Right? Maybe wrong! AI has <a href="https://en.wikipedia.org/wiki/Generative_adversarial_network">Generative Adversarial Network</a> (GAN) and Creative Adversarial Networks (CAN) have proven to generate art and sometimes better than humans. We will discuss this in some other article.</p><h4>References</h4><ol><li>pix2code: Generating Code from a Graphical User Interface Screenshot by Tony Beltramelli <a href="https://arxiv.org/pdf/1705.07962.pdf">https://arxiv.org/pdf/1705.07962.pdf</a></li><li>Deep Visual-Semantic Alignments for Generating Image Descriptions by Andrej Karpathy, Li Fei-Fei <a href="http://cs.stanford.edu/people/karpathy/cvpr2015.pdf">http://cs.stanford.edu/people/karpathy/cvpr2015.pdf</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=100deb28f54d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Blockchain- (Explain Like I’m 5)]]></title>
            <link>https://medium.com/@vijay.betigiri/blockchain-explained-like-im-5-yrs-5f04b91b059c?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/5f04b91b059c</guid>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[explain-to-kids]]></category>
            <category><![CDATA[ethereum]]></category>
            <category><![CDATA[distributed-systems]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Fri, 22 Dec 2017 10:32:41 GMT</pubDate>
            <atom:updated>2017-12-25T08:57:03.121Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Because of sudden surge and volatility in bitcoin prices, there is a lot of interest in blockchain (the technology behind bitcoin and other cryptocurrencies). But understanding blockchain is a little overwhelming even for IT nerds. Blockchain is significant not only for financial transactions, but also for any B2B/B2C/C2C transaction and thus it can disrupt most of the future businesses. I will try to explain it in easiest way.</em></p><h4>A Barter System</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VvmIpr1McKUNdZplF9efuA.png" /><figcaption>A teacher wanted to teach trading for school kids like a barter system</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lEx_USfY4Kjhh00qaNgzVA.png" /><figcaption>All kids were instructed to make a note of their trade.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XAtcxp4mQFVzTHqgTLsMDw.png" /><figcaption>But Reya promised to give one pencil to 2 friends and later erased her one note</figcaption></figure><h4>Centralized Ledger System (e.g. Bank)</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-VXd4qv4dVTs33w5T6isSQ.png" /><figcaption>Tom complained to her teacher about Reya. Teacher decided to be the 3rd party admin (like a bank).</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FDuugobQBO7lFO8fYXKTMg.png" /><figcaption>Things were going smoothly with centralized ledger system, until complaints started pouring…</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H7PGo1I1O-X_I1soQ-l_jQ.png" /><figcaption>In digital world, digital signature is replacement of physical signature.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PSSvxMVUg6pUwIKY8WLUhQ.png" /><figcaption>There is still a risk of tampering with the data.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jk8GzOyNiwCYspYODInxxQ.png" /><figcaption>The records can’t be traced back to previous transactions. Some sort of linking is required.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2XL2abfI0mx2iELevVw-3w.png" /><figcaption>Third party (such as bank) can not be trusted completely and they are redundant.</figcaption></figure><h4>Blockchain based Distributed Ledger</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sDg40T3uQz394BtVcMhZ4g.png" /><figcaption>Distributed data storage makes the centralized control with the third party redundant</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BH4SXWNODnFaZyhLhCWbpw.png" /><figcaption>In heavy concurrent transactions, it is difficult to maintain data integrity.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7zyxTjvjzzMpoLOdKt5zQg.png" /><figcaption>Volunteers (read miners) does the job of checking new transactions and add it as record.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1r-SxcEa8JQwQ6u2NAV5TA.png" /><figcaption>The volunteers (read miners) compete in calculating a magical number (read nounce). The winner gets the chance to add a new record.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K5Cj0VsWQ8EcHO_8vVWJKw.png" /><figcaption>Hash is a function of nounce and other data. So any change in magical number will result in completely new hash number.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pY2bdmkbpOnHao3OIaaMbQ.png" /><figcaption>Change in hash which doesn’t follow defined rules will be labeled as invalid block.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VGtpCXB-ihM0J0klAc1o1Q.png" /><figcaption>Hash identifies validity of block. Nounce helps in identifying the invalid block easily. Previous hash value links the blocks.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ic15Fcld2isIGQcgYdTaig.png" /><figcaption>Blocks are chained and hence the name blockchain.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-Hs3GsmNz_V6BrBejV9G7w.png" /><figcaption>Use of nounce value is to make the hash unique in some ways, so that any changes are caught quickly.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LZwJAf9-U4ukkJi95pYxxA.png" /><figcaption>Chain effect of blockchain</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e2R94x7WKFZRcAUh4q6FcA.png" /><figcaption>Blockchain is highly effective where there is no trust on other party, such as business!</figcaption></figure><p>You can read volunteers as miners, magical number as nounce value.</p><p>In a nutshell, blockchain provides 3 benefits-</p><ol><li>Eliminate third party admin/ regulators and still enables to do transactions between mutually distrusting parties</li><li>Provide a transparent chain of records</li><li>Avoid data tampering/ fraud</li></ol><p>We have covered a significant ground about blockchain in this explanation. <br>If you have liked the explanation, please clap for me. Thanks!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5f04b91b059c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deep Neural Network — (Explain Like I’m Five)]]></title>
            <link>https://medium.com/@vijay.betigiri/deep-neural-network-explain-like-im-five-6592e9c19a8c?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/6592e9c19a8c</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[explain-to-kids]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[deep-neural-networks]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Wed, 20 Dec 2017 16:08:11 GMT</pubDate>
            <atom:updated>2017-12-25T08:24:22.037Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Deep Neural Network (and more generally machine learning) is the most highly sought after technology skill, as it is going to change our lives more than what we can imagine. But learning Deep Neural Network can become overwhelming for beginners. I have attempted below to explain deep neural network in the easiest way with an analogy.</em></p><p>I had a favorite cook, who used to prepare two dishes. He was my favorite cook, as he knew what dish to prepare to make me happy.</p><p>But he kept his rules secret. He used to see at the inputs (on the left side of the image) such as sunny, rainy or guests and decide to prepare pizza or cake (shown on the right side).</p><p>Apparently his secret rules were,<br>1. Prepare pizza on rainy days<br>2. Cake on sunny days <br>3. Cake if guests are visiting, even on rainy days</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aYzQhkTccjY7RpCkFurKcg.png" /></figure><p>But one day he left the job!</p><p>New cook joined who had no idea about the secret rules.<br> I told him that I have high expectations from him.<br>The smart cook asked for few days of training time and my feedback every day during the training time.</p><p>I agreed!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/792/1*J7uLK_DVizc9om97ukVlmg.png" /></figure><p>On day 1, he prepared a pizza on a sunny day.</p><p>I shown my displeasure by saying that I would have preferred something sweet. He took a note of it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2vUXLyO_5KDDtyVrWKdXbQ.png" /></figure><p>On day 2, he prepared a cake when it was raining outside.</p><p>I told him that pizza would have suited my mood Today. He took a note of preparing pizza when it will be raining.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*POCH1q-FpjxxEbkq2jIKqg.png" /></figure><p>On day 3, he prepared a pizza as it was raining outside. My friends were visiting at my house.</p><p>I told the cook that my friends like cake more than the pizza, whatever is the outside condition. He noted of the new rule.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yQbz1Sk8dzG0amMsouOlkw.png" /></figure><p>On day 4, the cook declared that he is trained now and can correctly decide the dish now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OaVOg66HgXLRN9t4hbwpjQ.png" /></figure><p>Neural networks are similar to the smart cook.</p><p><strong>Neural Networks are good at duplicating any behavior through trial and error.</strong></p><p>Neural network communicates using numbers. If input is a true, it will be called as 1 and 0 if false. <br>Blue circles below are called as nodes. The nodes are connected by neural connections with a weight. More the weight, stronger is a the connection. Think of weight like a number of votes that the node is voting for. The vote can be positive or negative. The votes at the final layer are added to decide the choice.</p><p>Day 1 trial is the iteration 1 of training. Below is the connection assigned randomly. Since raining and guest inputs are zero, their nodes do not contribute in the voting process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EajDbtLZjYIP6gZLn-gIDQ.png" /></figure><p>As feedback is not okay, the neural network corrects the weight (vote) as,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mHJ5PzpiM6cgOReUHiB-pQ.png" /></figure><p>On iteration 2, neural network assumes the following,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IyaTFLvXrrTyct6OGaUMfg.png" /></figure><p>After learning from the feedback, the neural network corrects the weights (votes) as below,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9wmid0SHZDwWDNT6wa4hhQ.png" /></figure><p>Iteration 3: The neural network has already learnt about pizza when raining so it votes for pizza.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ja6SRORW-kIn2aQd_q6GBg.png" /></figure><p>But the cake was expected for friends so the neural network modifies the guest related votes to give more votes to cake than the pizza. The layer on the right does sum of the votes to derive at cake as the expected output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IVUug7iJXyg6iZEhsdt0Tw.png" /></figure><p>After the training phase, the neural network is a combination of all the learnt connections. Connections with 0 weight (vote) are omitted in the diagram as they don’t contribute in the voting process. Anytime a input is 1 at input (say sunny day), the node will generate it’s output (cake).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tCdHtWGwTXKwp5P5EJUL4A.png" /></figure><h4>Let’s go DEEP</h4><p>My choices over time have become complex. My changed preferences are as below,</p><ul><li>If there is a party at home with friends and parents then I want a cake.</li><li>Sometime, I do cost cutting when lot of guests at home and too many veggies in fridge. I prefer pizza in those times.</li><li>If I am playing cricket on a bright day with my friends, then I prefer a cake.</li><li>While I watch football on TV, I like to eat pizza. I watch football match when I have friends at home and can’t go out to play because of rain. Also parents should not be at home.</li><li>If there is a party at home and I am low on budget, then I will have pizza instead of cake.</li><li>I do study when it’s bright outside and no-one at home. I eat pizza on these days.</li></ul><p>We can work with these conditions similarly to the above story and derive at the below trained ‘DEEP’ neural network,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J6CF-wUK0j5CGQTrRtDLCg.png" /></figure><p>The red line indicates negative weight and usually it is -1. Exception is -3 and -2 weights shown above. All the green lines are carrying the positive weight of +1. These weights (votes) are added at the final layer to derive at the choice of dish.</p><p>The above neural network is called as Deep Neural network as, it has a input layer on the left, output layer on the right, and (at least) 1 hidden layer in the middle. In complex modeling, the hidden layer numbers may increase.</p><p>This was a binary classification problem, as the output had only two values (cake, pizza). Other examples of binary classification can be yes/no, present/absent, good/ defective, cat/no cat. We can extend the above problem to multiple choices of dishes (say cake, pizza, burger, sandwich), then it becomes a multiclass classification problem.</p><p>If you have liked the blog, please clap for me. Thanks!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6592e9c19a8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Artificial Intelligence (AI) driven Infrastructure]]></title>
            <link>https://medium.com/@vijay.betigiri/ai-defined-enterprise-it-a662b4fd9ca8?source=rss-6139f9655848------2</link>
            <guid isPermaLink="false">https://medium.com/p/a662b4fd9ca8</guid>
            <category><![CDATA[infrastructure]]></category>
            <category><![CDATA[digital]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[infrastructure-as-code]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Vijay Betigiri]]></dc:creator>
            <pubDate>Sun, 17 Dec 2017 10:58:50 GMT</pubDate>
            <atom:updated>2018-10-11T07:18:26.606Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/690/0*SGj0Gr272LEABR6L.jpg" /></figure><p><em>With developments in Cloud Computing and Software Defined Infrastructure, IT infrastructure has become on-demand, abstract and flexible. But still infrastructure is not intelligent. It is dependent on human intervention to understand the relationship between different IT components, understand the data trends and act accordingly. AI can change this and lead to error-free, self driven infrastructure.</em></p><h4>Role of cloud computing in Infrastructure</h4><p>Thanks to cloud computing, we are living in a world where the challenges of on-demand infrastructure is a past memory. Cloud computing has transformed every area of Enterprise IT, but the impact has been the most significant in the area of infrastructure. This has been possible because of cloud with its huge infrastructure capacity and automation has broken the ceiling of most quality benchmarks such as scalability, availability, elasticity, and performance.</p><h4>Software Defined Data-Center</h4><p>Software Defined Data-Center (SDDC) is leading to better infrastructure flexibility by pooling, aggregating and delivering the infrastructure capabilities as a software code. A SDI (Software Defined Infrastructure) abstracts the software from the hardware layer. Thus, the intelligence is moved from hardware components to a software stack. Based on software and a high degree of automation, a SDI is designed to build and control an infrastructure mostly without human interaction.</p><p>Recently I have worked on building a cloud-based platform, where all the possible infrastructure management (provision, scaling, user management etc.) is abstracted into a handful of API calls. These APIs can be consumed from the software itself or modern UI channels such as facebook/ slack chat or voice-based Alexa. Under the hood lies the complexities of multiple cloud environments, integration, services, security, monitoring, self-healing environments etc.</p><p>Thus a SDI-based infrastructure works independently from a hardware environment. However, a SDI is everything but intelligent! It is based on developer-defined rules (based on user’s previous experience) into hard-coded commands for taking certain actions automatically.</p><h4>AI Defined Infrastructure (AIDI)</h4><p>Enterprise AI applications are expected to grow at a compound annual growth rate (CAGR) of 64.3% in the 10 years (from $358 million in 2016 to $31.2 billion by 2025). Thus it is highly essential for IT and business to take early strides in this direction.</p><p>AI is defined in many ways. But in the current context, the intelligence of AI means a system which can self-learn the relationship between different IT components, understand the data pattern and the context, and act accordingly. The learning of its own environment makes a difference between the SDI and AIDI.</p><p>AI systems are dependent on large structured/ unstructured data which becomes a challenge in most of the AI based implementation. One huge advantage that infrastructure has is the availability of a lot of data in terms of user requests/ complaints, system logs etc.</p><h4>Use cases of AI defined Infrastructure</h4><p>Without any human intervention, AIDI can manage following tasks,</p><p>Planning:</p><ol><li>Analyze the demand trends and predict the infrastructure requirements and plan accordingly.</li><li>Match the requirements with the available infrastructure.</li></ol><p>Build:</p><ol><li>Deploy the necessary resources as per workload requirements.</li><li>De-allocating the resources when they are not needed anymore.</li><li>Configure the infrastructure components.</li></ol><p>Run &amp; Maintain:</p><ol><li>Analyze the data pattern which indicates the behavior of the system to make a model of the quality parameters w.r.t. system behavior. There is no need to define the rules, set threshold, as AI uses training to build their own model.</li><li>Use the model to achieve the most optimum quality parameters such as availability, scalability, storage.</li><li>Know about anomalies by knowing how the system behavior doesn’t look like. Identify anomalies such as intrusion detection, fraud/ fault points and save the infrastructure from abuse or going down.</li><li>In situations like cyber attack, where the gap between identification and mitigation is very important, AI can detect the threat and act immediately. We are already using a similar system, which checks your mail and detects if it contains anything suspicious and stops it from harming your system.</li><li>React or proactively act based on the single/group of infrastructure components. Autonomously take action to enable error-free infrastructure.</li><li>Reduce the cost of IT infrastructure by using the most optimal components. (e.g. instead of using 3 small server instances, choose to operate with 1 medium instance).</li></ol><h4>Improving the AI system</h4><p>Though designing AI system based on past system data can be helpful, but additional improvement can be brought by,</p><ol><li>Consume the knowledge from the experts, best practices to become more intelligent.</li><li>Learn by connecting with other AI systems. New working models can be easily built by techniques such as transfer learning from existing matured AI system (either from the same or different domain).</li><li>Never stop learning from new knowledge pool.</li><li>Optimize the knowledge as per business priorities.</li></ol><h4>Other ways AI is interacting with Infrastructure</h4><p>Apart from being the savior for infrastructure management, there are some interesting ways in which AI is impacting the infrastructure.</p><ol><li>Increased demand for infrastructure. AI requires huge, parallel computing power (such as GPUs) and large storage capacity.</li><li>The SDLC for Machine Learning is different from traditional application development. With machine learning, the development may happen on local setup, but training is recommended on cloud based GPUs. There is a lot of switching required between these environments. A good DevOps is essential considering this.</li><li>There are some black hats as well who will use AI into their job of threatening the IT. There will be an increase in disruptions and loss because of AI-based cyberattacks.</li><li>Infrastructure helpdesk support is currently manned by various levels of IT support executives. AI can take over the role of supporting their users 24 x 7 x 365 with a smart assistant system. Continuous improvements in Natural Language Processing and Sentiment Analysis is promising such systems.</li><li>Reduce data-center electricity usage significantly by using AI systems. Google has observed reduction by 40% of such cost by using AI.</li><li>Use robotics guided by AI to handle operations which needs hand-and-feet support.</li></ol><h4>Is it Easy to implement AI based Infrastructure?</h4><p>If AI can learn itself without needing to code, it must be very easy to develop such AI systems, right? Wrong! Though AI doesn’t need much of programming for building the model, data scientists need to choose machine learning algorithm or design a deep neural network. Also, a lot of considerations needs to be given in choosing quality data, algorithms and hyperparameters that define the design of neural networks.</p><p>Unlike rule-based traditional systems, AI systems can be error-prone as they have learned from the past data and the new data may be completely defying their model logic. A good domain knowledge is highly essential for improving the accuracy of the AI-based systems.</p><h4>AI defined Enterprise Architecture</h4><p>In this post, we have discussed how AI defined infrastructure enables IT in changing the infrastructure behavior from a today’s semi-dynamic to a true real-time IT environment in an automated way.</p><p>This autonomous way of planning, building, running and maintaining the entire infrastructure let IT operations and developers deploy IT resources like a server, storage, network, databases and other ready services in the most efficient way — by using the knowledge of experts and implementing intelligence based on the past data. Though this is not limited to infrastructure alone but have the ability to transform the end-to-end Enterprise Architecture. We will discuss this in the future blog of AI defined Enterprise Architecture.</p><p><strong>References:</strong></p><ul><li><a href="https://www.cio.com/article/3234656/artificial-intelligence/ai-defined-infrastructure-be-prepared-for-a-new-generation-of-business-models-and-applications.html">AI-defined infrastructure: be prepared for a new generation of business models and applications</a></li><li><a href="https://www.cio.com/article/3227608/artificial-intelligence/introducing-the-ai-defined-infrastructure.html">Introducing the AI-defined infrastructure</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a662b4fd9ca8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>