What I learnt in the 5-Day Gen AI Intensive Course with Google

This is a submission for the Google AI Agents Writing Challenge: Learning Reflections

Hello!! I recently completed the 5-Day Gen AI Intensive Course with Google, mainly out of 2 reasons, first, I like Python too much( sorry Typescript :( ) and secondly, AI Agents are cool!! So, I though why not make a post sharing what I learnt in this, along with how I have developed some AI Agents before.

So, let me take you through the days,

November 10th, 2025. I signed up for the AI Agents Intensive with a portfolio of agent projects already under my belt—Discord bots that answered documentation queries using RAG, a real-time voice agent built with LiveKit and FastAPI, even a customer support system for a hackathon. But here's the truth: every single project felt like I was building a house of cards. One weird edge case, one hallucinated API call, one unexpected user input—and everything collapsed. I would vibe-code a little bit more than my understanding, not taking the fundamentals in mind, so, I used NotebookLM and wanted to understand the whitepaper on foundational LLMs explained transformers from first principles, and suddenly I understood why my prompts worked or failed. When I completed the codelab on prompting fundamentals, I learned about parameters I'd been ignoring—temperature, top-k, top-p—and how they actually controlled output randomness and creativity.

The evaluation codelab changed everything. I'd been manually checking agent outputs like a caveman. Now I learned about autoraters and structured outputs—ways to programmatically verify if my agent was hallucinating or staying on track. Pair it with the embeddings whitepaper finally explained what I'd been cargo-culting. I'd used Chroma and vector stores before, but never understood why certain searches failed. Building the document Q&A system with RAG from scratch taught me chunk sizing, overlap strategies, and how embedding quality directly impacts retrieval.

So, the whitepaper's were quite high level for me, as I know some basic ML/DL, but I tried to understand the low lying levels through Gemini(ahh yes!! understanding AI using AI), but then as I though the AI couldn't get any smarter, the next I got to know about multi-agents(I implemented this in 1 of my projects maybe, slight of).I understood state machines, agent loops, and how to handle multi-turn conversations without losing context.

Domain-Specific LLMs are something I like, it is more relevant than working with an LLM that knows a little bit of everything, and I personally like those. I now understood more and more, as I filled the knowledge gaps.

But as all things come to an end, so does this course, MLOps, hmm do people like joining Ops with everything? DevOps, DataOps, what next? So, Day 5, I understood and watched the video, as there was no codelab, although it would have been better if we could make a sandbox and deploy the agent, but I enjoyed the course and would like to end by saying AI Agents are quite interesting and fascinating, and I think they will be more and more used by people, helping them in everyday lives, making their lives easier.

So, this was all from me today!! Hope you liked it, comment down what you liked/disliked about my post or the AI Agents course. See ya!!