The latest articles on DEV Community by Shankar Chavan (@sc4685215). https://dev.to/sc4685215 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fthepracticaldev.s3.amazonaws.com%2Fi%2F99mvlsfu5tfj9m7ku25d.png https://dev.to/sc4685215 en Shankar Chavan Thu, 06 Aug 2026 16:25:51 +0000 https://dev.to/sc4685215/my-gen-ai-academy-cohort-3-journey-from-zero-to-deployed-agent-4fj0 https://dev.to/sc4685215/my-gen-ai-academy-cohort-3-journey-from-zero-to-deployed-agent-4fj0 <h1> My Gen AI Academy Cohort 3 Journey: From Zero to Deployed Agent </h1> <p>When I signed up for the Google Cloud Gen AI Academy APAC Edition, I wasn't just looking for another certificate to add to my resume. As a final-year Computer Engineering student, I wanted hands-on experience actually deploying an AI agent — not just reading about how one works.</p> <p>That's how StudyMate came to life.</p> <h2> What I Built </h2> <p>StudyMate is an AI agent designed to help students study smarter, deployed on <strong>Google Cloud Run</strong>. The idea was simple: give students a personalized assistant that can answer questions, organize study material, and adapt to how they actually learn — available anytime, without needing a beefy local setup.</p> <p>I built it using the Agent Development Kit (ADK) and grounded it with Retrieval-Augmented Generation (RAG), so instead of hallucinating answers, StudyMate pulls from real study content to give grounded, relevant responses. Then I containerized and deployed it on Cloud Run, which meant I didn't have to worry about managing servers — I could focus entirely on making the agent actually useful.</p> <h2> The Challenge That Taught Me the Most </h2> <p>The trickiest part wasn't the AI logic — it was getting the RAG pipeline to retrieve the <em>right</em> context consistently. Early versions of StudyMate would sometimes answer confidently using the wrong source material. Debugging this taught me more about prompt engineering and retrieval tuning than any tutorial could have. It forced me to actually understand what's happening between a user's query and the model's response, instead of treating RAG as a black box.</p> <h2> Why This Matters to Me </h2> <p>I run an AI-animated YouTube channel and tinker with side projects like MarketMind, a multi-agent LLM stock sentiment simulator — so I'm no stranger to building with AI. But the Gen AI Academy pushed me to think about <em>deployment</em> and <em>production-readiness</em> in a way my personal projects hadn't. Cloud Run specifically showed me how lightweight and scalable a well-built agent can be, without the overhead of managing infrastructure.</p> <h2> What's Next </h2> <p>I'm planning to extend StudyMate with better memory across sessions and explore multi-agent collaboration — inspired by some of the same architecture patterns I've been experimenting with in MarketMind. The Academy gave me a strong foundation in ADK, RAG, and Cloud Run, and I'm excited to keep building on it through the rest of Cohort 3.</p> <p>If you're a fellow builder in the Gen AI Academy, I'd love to hear what you're building too — drop a comment below!</p> webdev devops cloud ai