<?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 Agustinus Nalwan on Medium]]></title>
        <description><![CDATA[Stories by Agustinus Nalwan on Medium]]></description>
        <link>https://medium.com/@agustinus-nalwan?source=rss-8b7ab157b0a4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*Y3IOQ8vHmCTDwP2NQGwOSw.jpeg</url>
            <title>Stories by Agustinus Nalwan on Medium</title>
            <link>https://medium.com/@agustinus-nalwan?source=rss-8b7ab157b0a4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 14:40:10 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@agustinus-nalwan/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 Real Impact of Generative AI: Beyond the Hype Cycle]]></title>
            <link>https://agustinus-nalwan.medium.com/the-real-impact-of-generative-ai-beyond-the-hype-cycle-4ac17eab4f99?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/4ac17eab4f99</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Fri, 11 Oct 2024 03:48:26 GMT</pubDate>
            <atom:updated>2024-10-11T03:48:26.038Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why You Should Act Now to Leverage Generative AI for Business Growth and Stay Ahead of Competitors</h4><h3>Introduction</h3><p>The release of ChatGPT in December 2022 ignited a boom in Generative AI (Gen AI). This breakthrough spurred a race among companies to develop increasingly advanced large language models (LLMs), including multimodal ones. Startups have flourished, building products with enhanced capabilities leveraging these LLMs. Major firms – such as NVIDIA, cloud providers, and even phone and laptop manufacturers – are pouring investments into Gen AI to avoid missing early adopters who could become future customers. Industries across the board are keen to capture this wave of innovation.</p><h3>The Gartner Hype Cycle and Its Dangers</h3><p>With any new technology, there are believers and skeptics. The Gartner Hype Cycle suggests that new tech undergoes inflated expectations before plummeting into the “trough of disillusionment.” Many believe that Gen AI has just passed its peak of inflated expectations and is poised for a fall. While the Hype Cycle accurately describes many new technologies, focusing solely on it can be misleading.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pIwdiKwYl962s3eU4m09lA.jpeg" /><figcaption>Gartner Hype Cycle — Where Gen AI is now</figcaption></figure><p>Exaggerated claims often stem from enthusiasts, developers who haven’t fully assessed limitations beyond prototype use cases, or vendors eager to promote their platforms, a group I called hyper-inflater. Consequently, organisations attempting similar feat to build production-grade systems may encounter challenges, leading to disillusionment.</p><h3>The Risk of Staying on the Sidelines</h3><p>This blog offers a different perspective: the danger for organisations that, fearing over investment due to the Hype Cycle, choose to wait and stay on the sidelines — the skeptics. This groups expectation of Gen AI capabilities is represented by the grey lines (skeptic fallacy) — very low expectations and eventually as many organisation sharing success stories it started to catch up at the plateau of productivity point.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z1zrUgiyuDr1b_4shCkIlg.jpeg" /><figcaption>Skeptic fallacy and real capability curve</figcaption></figure><p>The thing is… not all capabilities are hype; a significant portion is real, offering tangible benefits that will improve over time. Gen AI’s real capabilities, represented by the green curve are substantial enough to uplift business operations by unlocking new ideas, enhancing processes, boosting team efficiency, and improving customer experiences.</p><p><strong><em>More importantly, these capabilities are potent enough to disrupt and threaten businesses, even from smaller competitors who effectively harness them.</em></strong></p><p>Organisations that over-rely on the Hype Cycle may miss out on leveraging Gen AI’s true potential, finding themselves too late if competitors have already capitalised on these technologies which is represented by the difference between missed out opportunity gap (real capability minus their existing non Gen AI capability)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ej2zRlBK4uHumT43_HjSIw.jpeg" /><figcaption>Missed out opportunity</figcaption></figure><p>If you’re among those hesitant to invest in Gen AI, this blog aims to encourage you to reconsider and focus on the real capabilities you can leverage now.</p><h3>My Experience with Gen AI</h3><p>To avoid being labeled as a hyper-inflater, let me share my experience where I based my information in this blog from. I lead an AI division that has successfully built and deployed three LLM-powered consumer-facing technologies, including Retrieval Augmented Generation (RAG). As part of our AI Governance Committee, I rigorously review AI developments, especially those powered by Gen AI, ensuring safety and responsibility. Our organisation uses ChatGPT’s team license, which has demonstrably improved employee productivity.</p><p>In the rest of this blog, I’ll share my experience how Gen AI has already delivered tangible benefits in four areas.</p><h3>Gen AI Accelerates AI Development</h3><p>Gen AI is replacing many traditional NLP AI techniques by offering better accuracy, shorter development times, and significantly reduced maintenance. For example, consider detecting popular topics and sentiments from customer product reviews.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Mknzz18RR-SoUc1U" /><figcaption>Generative AI simplify many NLP tasks (Image generated with MidJourney)</figcaption></figure><p>Traditionally, a data scientist would spend weeks collecting and cleaning data, splitting reviews into paragraphs or sentences, labelling each chunk, training models, and tuning hyper-parameters. An ML engineer would then spend additional weeks productionising the model, building continuous training pipelines, and maintaining the system.</p><p>With Gen AI, specifically LLMs, this process is streamlined and can be at least three times cheaper:</p><ul><li><strong>Data Collection and Labelling — </strong>Data scientists still collect and label datasets but primarily to evaluate prompts against an LLM.</li><li><strong>Prompt Engineering</strong> — They perform prompt engineering iterations, evaluating prompts against validation sets until acceptable accuracy is achieved.</li><li><strong>Simplified Production</strong> — Productionising the model becomes much simpler – essentially an API call to an LLM provider. There’s no need to handle large model binaries or complex ML frameworks, significantly reducing maintenance and infrastructure costs.</li></ul><p>Often, the LLM solution provides higher accuracy, as it can process entire reviews without losing context – a common challenge with traditional NLP techniques.</p><p>In the past six months, LLMs have become our first choice for many NLP tasks like sentiment analysis, entity extraction (e.g., detecting brand names, locations), and text classification. This shift has tremendously boosted our productivity.</p><h4><strong>Addressing Common Concerns</strong></h4><ul><li><strong>Cost</strong> — While some use cases of LLMs can be expensive, various sizes are available, and costs are proportional to size. You should evaluate and choose the smallest model that meets your acceptable accuracy. When factoring in labor costs for building, productionising, and maintenance, the LLM option is often way more cost-effective.</li><li><strong>Speed </strong>— LLMs can be slow, but selecting the appropriate size can improve response times. Moreover, for applications like sentiment analysis of customer reviews, real-time processing isn’t always necessary.</li></ul><p>With the extra time saved, your team can support your business better by building more AI technologies. Similarly, your competitors who are leveraging this capability may be building AI solutions at three times the speed.</p><h3>Enhancing Customer Experience with Gen AI</h3><p>Gen AI doesn’t just improve AI models’ accuracy; it also enables you to deliver highly personalised web or app experiences. For instance:</p><ul><li><strong>Amazon </strong>— introduced Rufus, a chatbot that answers user questions about specific products and even compares items like running and hiking shoes.</li><li><strong>Carsales </strong>— implemented an <a href="https://medium.com/towards-data-science/enhancing-direct-answer-accuracy-in-rag-setup-with-self-retrieval-mechanisms-8162c8cc7853">AI search feature</a> for automotive editorial articles, returning relevant content and direct answers when applicable. This resulted in increased click-through rates and positive user feedback.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pkFk_kabp1o7oqEoVKI98A.jpeg" /><figcaption>carsales.com</figcaption></figure><h3>Empowering Software Engineers with AI Skills</h3><p>In many organisations, software engineers outnumber data scientists and ML engineers. Some organisations might not have any data scientists or ML engineers at all, limiting their ability to build AI technologies.</p><p>I’m a strong advocate for AI democratisation. While software engineers can be trained to become citizen data scientists or ML engineers, barriers exist – many may not be proficient in Python and ML techniques, for example.</p><p>With Gen AI, many LLM use cases involve simple API calls accessible through frameworks available in most programming languages. This accessibility allows software engineers to leverage AI capabilities with proper guidelines on evaluating LLM models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*d4w7ifGId0dmuw0i" /><figcaption>Generative AI is enabling many software developer (Image generated with MidJourney)</figcaption></figure><p>Imagine the immense capabilities and efficiency gains when all your software engineers can leverage AI. Similarly, your competitors might already be building AI tech more cheaply by utilising their existing software engineers.</p><h3>Boosting Non-Technical Team Productivity</h3><p>Gen AI significantly boosts productivity for non-technical teams as well. While many productivity tools like Jira, Microsoft Office Suite, Confluence, and Miro are integrating Gen AI capabilities, standalone chatbot tools like ChatGPT, Claude for Work, or Perplexity.ai are highly useful. They help team members perform tasks such as:</p><ul><li><strong>Summarising Long Documents</strong> — Quickly distill essential information from lengthy reports.</li><li><strong>Brainstorming Ideas </strong>— Generate creative concepts or solutions to problems.</li><li><strong>Analysing Large Datasets </strong>— Examine extensive data like product reviews without waiting for specialised teams.</li><li><strong>Paraphrasing Documents </strong>— Reword content for clarity or different audiences.</li><li><strong>Conducting Market Research </strong>— Gather insights on market trends and competitors efficiently.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*p9LQJG0SXK0UdXq3" /><figcaption>Gen AI is helping non technical team (Image generated with MidJourney)</figcaption></figure><p>Our test group have reported an over 5% productivity increase by using tools like ChatGPT. For a $20 monthly license, this offers a high return on investment for businesses. Tasks like analysing large datasets no longer require creating support tickets and waiting for data science or analytics teams, significantly improving efficiency.</p><p>However, guidelines are necessary to ensure teams know how to validate outputs and practice effective prompt engineering. The last thing you want is a business decision based on inaccurate analysis produced by an LLM.</p><p>Again, consider that your competitors leveraging these tools may have over 5% productivity boost.</p><h3>Summary</h3><p>Gen AI is not just hype; it’s already delivering real business value. While exaggerated claims are common, ignoring its potential could leave organisations at risk. Gen AI accelerates AI development, enhances customer experiences, and boosts productivity by over 5%, even allowing software engineers to build AI solutions faster. The pace of innovation is swift, with advancements doubling every three months (GPT4o is now 50x cheaper compare to original GPT4 in March 2023). Testing Gen AI with low-risk pilots is the best way to understand its potential.</p><p>Caution is important — while many use cases may not be direct customer-facing and thus have less emphasis on safety, customer-facing chatbots do require strict safety measures to prevent reputation damage when providing misleading answer. Establishing strong AI governance and data validation practices can help mitigate these risks, which I will cover in a future blog. However, the risk of inaction is greater, as staying on the sidelines may leave your business vulnerable to competitors already leveraging Gen AI’s powerful capabilities. Balancing caution with innovation is key to thriving in this rapidly evolving landscape.</p><p>If you enjoyed this blog, please show your support with plenty of claps.</p><p>To connect with me, feel free to add me on <a href="https://www.linkedin.com/in/agustinus-nalwan/">LinkedIn</a>. For more of my writings, you can follow me on <a href="https://medium.com/@agustinus-nalwan">Medium</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4ac17eab4f99" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building skAi, The Ultimate Goal Oriented Smart Home Concierge AI Powered by Multi-Modal LLM Agent]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/swlh/building-skai-the-ultimate-goal-oriented-smart-home-concierge-ai-powered-by-multi-modal-llm-agent-7ac6cb002972?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/968/1*VNW9WZNNms6ev-x8fkrFUQ.png" width="968"></a></p><p class="medium-feed-snippet">As a follow up, we embarked on a week-long experiment living with the AI as a nanny, learning assistant, baby feeding notifier, and home&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/swlh/building-skai-the-ultimate-goal-oriented-smart-home-concierge-ai-powered-by-multi-modal-llm-agent-7ac6cb002972?source=rss-8b7ab157b0a4------2">Continue reading on The Startup »</a></p></div>]]></description>
            <link>https://medium.com/swlh/building-skai-the-ultimate-goal-oriented-smart-home-concierge-ai-powered-by-multi-modal-llm-agent-7ac6cb002972?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/7ac6cb002972</guid>
            <category><![CDATA[generative-ai-solution]]></category>
            <category><![CDATA[llm-agent]]></category>
            <category><![CDATA[multi-modal-llm]]></category>
            <category><![CDATA[claude-3]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Tue, 07 May 2024 08:27:35 GMT</pubDate>
            <atom:updated>2024-05-07T08:27:35.779Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Enhancing Direct Answer Accuracy in RAG Setup with Self-Retrieval Mechanisms]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/enhancing-direct-answer-accuracy-in-rag-setup-with-self-retrieval-mechanisms-8162c8cc7853?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1728/1*QsQ5IQVBznwnPNq88iOqdQ.jpeg" width="1728"></a></p><p class="medium-feed-snippet">Leveraging the power of LLMs to significantly enhance the quality of document context retrieved for direct answer generation in your RAG&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/data-science/enhancing-direct-answer-accuracy-in-rag-setup-with-self-retrieval-mechanisms-8162c8cc7853?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/enhancing-direct-answer-accuracy-in-rag-setup-with-self-retrieval-mechanisms-8162c8cc7853?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/8162c8cc7853</guid>
            <category><![CDATA[retrieval-augmented-gen]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[deep-dives]]></category>
            <category><![CDATA[rag-architecture]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Tue, 30 Apr 2024 16:51:50 GMT</pubDate>
            <atom:updated>2024-05-01T06:30:18.440Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Amazon Q: No Code RAG Solution that Auto Sync to Many Data Sources]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://aiadvances.org/amazon-q-no-code-rag-solution-that-auto-sync-to-many-data-sources-4958f281f9e6?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1430/1*sAxeuaQDLMhx5IhKe3V5XA.png" width="1430"></a></p><p class="medium-feed-snippet">Support ZenDesk, Confluence, S3, Microsoft Exchange, and many others.</p><p class="medium-feed-link"><a href="https://aiadvances.org/amazon-q-no-code-rag-solution-that-auto-sync-to-many-data-sources-4958f281f9e6?source=rss-8b7ab157b0a4------2">Continue reading on AI Advances »</a></p></div>]]></description>
            <link>https://aiadvances.org/amazon-q-no-code-rag-solution-that-auto-sync-to-many-data-sources-4958f281f9e6?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/4958f281f9e6</guid>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[amazon-q]]></category>
            <category><![CDATA[retrieval-generation]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Tue, 05 Dec 2023 08:48:13 GMT</pubDate>
            <atom:updated>2023-12-05T10:23:09.271Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Enhancing RAG’s Answer: Self-Debugging Techniques and Cognitive Load Reduction]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/enhancing-rags-answer-self-debugging-techniques-and-cognitive-load-reduction-8c4273013d39?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1344/0*qUrdsZIM-SQYVCXZ" width="1344"></a></p><p class="medium-feed-snippet">Asking the LLM to self-diagnose and self-correct the prompt to improve answer quality.</p><p class="medium-feed-link"><a href="https://medium.com/data-science/enhancing-rags-answer-self-debugging-techniques-and-cognitive-load-reduction-8c4273013d39?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/enhancing-rags-answer-self-debugging-techniques-and-cognitive-load-reduction-8c4273013d39?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c4273013d39</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-dives]]></category>
            <category><![CDATA[retrieval-generation]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Sun, 26 Nov 2023 15:07:51 GMT</pubDate>
            <atom:updated>2023-11-26T15:07:51.167Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Untold Side of RAG: Addressing Its Challenges in Domain-Specific Searches]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/the-untold-side-of-rag-addressing-its-challenges-in-domain-specific-searches-808956e3ecc8?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1834/1*Fb4Jq01ganD75PJDxHVknQ.jpeg" width="1834"></a></p><p class="medium-feed-snippet">Using hybrid search, hierarchical ranking, and instructor embedding to address similar domain-specific documents in our RAG setup</p><p class="medium-feed-link"><a href="https://medium.com/data-science/the-untold-side-of-rag-addressing-its-challenges-in-domain-specific-searches-808956e3ecc8?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/the-untold-side-of-rag-addressing-its-challenges-in-domain-specific-searches-808956e3ecc8?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/808956e3ecc8</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-dives]]></category>
            <category><![CDATA[retrieval-generation]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[document-embedding]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Wed, 18 Oct 2023 03:15:10 GMT</pubDate>
            <atom:updated>2023-10-21T04:57:03.093Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Building Owly an AI Comic Video Generator For My Son]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/building-owly-an-ai-comic-story-generator-for-my-son-c99fb695d83b?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1280/1*pTJZhPwZIJwVstlmcXLRAA.jpeg" width="1280"></a></p><p class="medium-feed-snippet">Utilising the fine-tuned Stable Diffusion 2.1 on Amazon SageMaker JumpStart, I developed an AI tech called Owly that crafts personalised&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/data-science/building-owly-an-ai-comic-story-generator-for-my-son-c99fb695d83b?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/building-owly-an-ai-comic-story-generator-for-my-son-c99fb695d83b?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/c99fb695d83b</guid>
            <category><![CDATA[ai-comic-video-generator]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Tue, 11 Apr 2023 02:44:20 GMT</pubDate>
            <atom:updated>2023-04-21T17:36:28.888Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Building Ellee — A GPT-3 and Computer Vision Powered Talking Robotic Teddy Bear With Human Level…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/building-ellee-a-gpt-3-and-computer-vision-powered-talking-robotic-teddy-bear-with-human-level-db7d08259583?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1690/1*P9HzTvupacWTND064FDmJg.jpeg" width="1690"></a></p><p class="medium-feed-snippet">Using an object detection and facial recognition system built on MobileNetSSDV2 and Dlib and running on an NVIDIA Jetson Nano, a GPT-3&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/data-science/building-ellee-a-gpt-3-and-computer-vision-powered-talking-robotic-teddy-bear-with-human-level-db7d08259583?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/building-ellee-a-gpt-3-and-computer-vision-powered-talking-robotic-teddy-bear-with-human-level-db7d08259583?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/db7d08259583</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-dives]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[gpt-3]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Fri, 04 Feb 2022 04:41:56 GMT</pubDate>
            <atom:updated>2022-02-08T15:17:02.107Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Staying in the tech — my career journey]]></title>
            <link>https://medium.com/carsales-dev/staying-in-the-tech-my-career-journey-de154c315cc3?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/de154c315cc3</guid>
            <category><![CDATA[career-development]]></category>
            <category><![CDATA[career-paths]]></category>
            <category><![CDATA[individual-contributor]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Wed, 01 Dec 2021 04:56:10 GMT</pubDate>
            <atom:updated>2021-12-01T04:56:10.860Z</atom:updated>
            <content:encoded><![CDATA[<h3>Staying in the tech — my career journey</h3><p>Career development and progression have many forms and are subjective to individuals. For some it can mean stepping up the ladder, leading a bigger team or focusing on being an individual contributor. Don’t copy others or expect others to tell you, but find which pathway fits you and that you enjoy doing the most. Here, I’d like to share my career progression journey with you and hopefully there are some learnings you can take out of it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/334/1*Xc20YEHn84kJx1Pvm2ZYRw.png" /><figcaption>Agustinus (Gus) Nalwan</figcaption></figure><p>I started my IT career prior to commencing a Bachelor’s degree in Computer Science and Robotics at Monash University, by building games and authoring a games programming book in 1995. Apart from this helping to financially support my tuition and living expenses as an overseas student, I found it very rewarding to solve hard problems and teaching others how to do so. In fact, the last 6 years of my career have reflected exactly this, just in different shapes.</p><p><strong>1998 — Games and 3D Computer Graphics Animation</strong></p><p>My first few jobs after graduation were in 3D Computer Graphics and Animation ranging from building a Motion Capture Software for Movie Special Effects called Famous3D and Flash Animation authoring tools called SwishMax.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*svQFvdxvg_FtGGY7m31l-w.png" /><figcaption>3D Motion Capture system</figcaption></figure><p><strong>2009 — iOS App Tech Lead</strong></p><p>I joined carsales.com ltd as their first and only iOS App Developer working in a team of five building carsales’ first iOS app. It was a web app (mobi site running in a WebView) and we had lots of problem trying to deliver the best experience under the constraint of not being a native app. My tenure did not last long. I resigned a year later due to the different alignment between the company’s vision and mine. I wanted to build the native iOS app but carsales wanted to keep building the web app. Three months later, I re-joined the same team with the new direction to build a native app.</p><p>We released our first app and our rating shot to 4.5 stars! In 2012, as the iOS team grew, I became the Lead iOS Developer, working together with and mentoring the junior team members. The apps teams itself was also growing, we grew an Android team, more API developers, a designer and a QA. As native app development was still not well understood, there was a lot of exploration and experimentation required which aligned to my interests in solving hard problems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*wQzk8OIEyq20SzMoKNf6kg.png" /><figcaption>Apps team celebrating app release</figcaption></figure><p><strong>2016 — AI/ML Tech Lead</strong></p><p>After a few years, as the native app development process became more mature and standard, I started seeking a new challenge in an uncharted territory. Coincidently, AI was the new hot topic that year. I raised my hand to take the opportunity to run the first AI guild at carsales, leading three teams of volunteers (developers from other teams who were given 1 day/week to participate in building AI tech in this guild) to build a POC to use AI to solve business problems. This initiative, give birth to Cyclops, image recognition which was proven to deliver business value in saving thousands of dollars automating the photographer’s process in classifying thousands of photos daily. This success gave birth to a formal small AI/ML team which I technically lead.</p><p>Since then, we have been building and productionising more and more AI tech powering various products within carsales to identify the car spec from photos, decoding the car spec from a vin number, automating approval of private ads and blurring rego numbers in a photo. Each of these projects comes with different challenges which was what I was/am after in my work.</p><p>We are a small team of super passionate individuals who all love to solve hard problems together and have so much fun at the same time. This is a small enough team which leaves plenty of opportunity for me to still be hands on in the technical.</p><p>One of the most memorable projects we did was deploying a car identification system on top of highways, identifying cars zooming pass at 100kph in real-time. A few times our team had to climbed up the highway at night to collect sample footage to train our model. The most challenging factor was that it was not feasible to climb the tower where the system was mounted to fix any bugs. The shear technical and logistic complexity of the project totally reset the bar of building a robust system with multiple layers of redundancy. It was nerve wrecking but very fun!</p><p>The first few years of this AI/ML chapter taught me one important lesson. It doesn’t matter how cool the technology you build is, if it doesn’t solve a problem, it won’t materialise as you will not get the support you need to integrate it into real products. I am still having fun building the technology with the team, however nothing compares to the feeling of seeing your technology being used by customers and to seeing the impact it brings. It is this that drives me further and encourages me to be passionate about the outcomes of our tech.</p><p>From here I started to think more strategically, investigating what were the problems our business was facing and trying to marry them with technology to solve these problems.</p><p><strong>2017 AI/ML Evangelist</strong></p><p>Realising how hard it was to build AI Tech in these early days, I started to speak at various AI &amp; Tech conferences to share my experience and knowledge. I also wanted to encourage others to jump into AI/ML and to grow the AI ecosystem in Australia which was far behind countries like the US and UK. Through this evangelism I was awarded an AWS AI/ML Hero in 2018.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*q2Tlt3F925ZtqmWER0EC1w.png" /><figcaption>Cyclops live demo at AWS Sydney Summit</figcaption></figure><p><strong>2020 Multiplier — Scaling Up AI &amp; MLOps</strong></p><p>We have been building and productionising AI tech since 2017. Today only two small teams within carsales are building AI models and only one team can productionise them. However, the demand for AI tech is growing rapidly. This is the common problem many organisations face. How do we scale up AI production without proportionally needing to grow the AI/ML team? I believe the answer is by enabling AI capability to all of our large pool of software developers with the help of tools and high level AI services.</p><p>So our AI team strategy shifted from building for others into helping others to build themselves. This is being achieved by building an MLOps platform and tools for other teams to use and providing training/guidance on AI.</p><p>You can be the best software engineer in the world, building a software 5x faster than normal engineer, however, you are still a human being and, there is still a maximum cap to your throughput. In order to surpass that cap you need to start influencing and multiplying and amplifying others around you, where you will bring much bigger impact to the business.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/904/1*UbxUb2_xq3dgKGzrgWfoTA.png" /><figcaption>AI Team running an AI workshops to spread AI knowledge within Carsales</figcaption></figure><p><strong>Summary</strong></p><p>As you can see from my early career days until now, my career has been about constant learning, solving challenges with similar minded people and encouraging and mentoring others to do the same. This is what career progression is for me.</p><p>Here are a few takeaways:</p><ul><li>Career is a lifetime journey and it is very important to enjoy what you do.</li><li>Take charge of your own career progression — Identify problems within your organisation which are aligned to your interests and raise your hand to propose solutions and lead the charge.</li><li>Money should not always be your biggest drive. It matters, but not if it means you don’t enjoy what you are doing every day.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=de154c315cc3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/carsales-dev/staying-in-the-tech-my-career-journey-de154c315cc3">Staying in the tech — my career journey</a> was originally published in <a href="https://medium.com/carsales-dev">carsales-dev</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Track, manage, discover and reuse AI models better using Amazon SageMaker Model Registry]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/track-manage-discover-and-reuse-ai-models-better-using-amazon-sagemaker-model-registry-a238ec73351c?source=rss-8b7ab157b0a4------2"><img src="https://cdn-images-1.medium.com/max/1000/0*Ndr0BFK97KZoJhgs" width="1000"></a></p><p class="medium-feed-snippet">Model registry is a centralised repository to store our models.</p><p class="medium-feed-link"><a href="https://medium.com/data-science/track-manage-discover-and-reuse-ai-models-better-using-amazon-sagemaker-model-registry-a238ec73351c?source=rss-8b7ab157b0a4------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/track-manage-discover-and-reuse-ai-models-better-using-amazon-sagemaker-model-registry-a238ec73351c?source=rss-8b7ab157b0a4------2</link>
            <guid isPermaLink="false">https://medium.com/p/a238ec73351c</guid>
            <category><![CDATA[model-registry]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[editors-pick]]></category>
            <category><![CDATA[tips-and-tricks]]></category>
            <category><![CDATA[mlops]]></category>
            <dc:creator><![CDATA[Agustinus Nalwan]]></dc:creator>
            <pubDate>Fri, 29 Oct 2021 03:51:11 GMT</pubDate>
            <atom:updated>2021-10-29T15:49:08.553Z</atom:updated>
        </item>
    </channel>
</rss>