Source: 5-Day Gen AI Intensive Course with Google — Kaggle Learn Guide (the original ran live 03/31–04/04/2025 and is now a self-paced guide). Day titles, whitepaper names, and speaker lists are copied from that page. The guide holds five days, each with a summary podcast, a whitepaper, and a few codelabs in Kaggle Notebooks. There is also a livestream Q&A with Google experts for every topic. A Kaggle account with phone verification runs the codelabs, and an AI Studio account issues the Gemini API key. The bonus notebook and the AI Agents Intensive guide sit outside the five days.
Overview
A free 5-day course run by Google in partnership with Kaggle, providing foundational knowledge of Generative AI. Each day includes: 1 summary podcast, 1 whitepaper for further reading, and (on most days) a few hands-on codelabs in Kaggle Notebooks; there is also a livestream Q&A with Google experts for each topic. It requires a Kaggle account (with phone verification to run codelabs) and an AI Studio account (to get a Gemini API key).
Day 1 — Foundational LLMs & Prompt Engineering
- Theory: the history of LLM development — from the transformer architecture to fine-tuning techniques, inference acceleration, and reasoning models. Whitepapers: "Foundational Large Language Models & Text Generation" and "Prompt Engineering".
- Codelabs: getting familiar with the Gemini API (how parameters affect the prompt), and how to evaluate LLM output with an autorater + structured output.
- Optional material: a case study of a bank using advanced prompt engineering to automate its financial advisory process; a recording of the livestream with Google speakers (Warren Barkley, Logan Kilpatrick, Kieran Milan, Anant Nawalgaria, Irina Sigler, Mat Velloso).
Day 2 — Embeddings and Vector Stores/Databases
- Theory: the conceptual foundations of embeddings and vector databases — how embeddings are generated, vector search algorithms, real-world applications combined with LLMs, and the tradeoffs between approaches. Whitepaper: "Embeddings and Vector Stores/Databases".
- Codelabs: build a RAG question-answering system over custom documents; explore text similarity through embeddings; build a Keras neural network classifier using embeddings.
- Optional: livestream with Andre Araujo, Patricia Florissi, Alan Li, Anant Nawalgaria, Xiaoqi Ren, Chuck Sugnet, Howard Zhou.
Day 3 — Generative AI Agents
- Theory: the core components of an AI agent, the iterative agent development process, advanced agent architectures (multi-agent systems), and how to evaluate agents. Main whitepaper: "Generative AI Agents"; advanced (optional) whitepaper: "Agents Companion".
- Codelabs: have a chatbot use function calling to "talk" to a database through a SQL tool (with an example using the Gemini 2.0 Live API); build a café order-taking agent with LangGraph.
- Optional: a case study of a regulatory reporting solutions provider using an agentic GenAI system to automate "ticket-to-code" in software development, achieving 2.5x productivity; livestream with Alan Blount, Antonio Gulli, Steven Johnson, Jaclyn Konzelmann, Patrick Marlow, Anant Nawalgaria, Julia Wiesinger.
Day 4 — Domain-Specific LLMs
- Theory: how to build and apply industry-specific LLMs such as SecLM (cybersecurity) and MedLM/Med-PaLM (healthcare), with talks from the researchers who created them. Whitepaper: "Solving Domain-Specific Problems Using LLMs".
- Codelabs: feed real-time data into the model through Google Search and then visualize it with a charting tool via the Live API; fine-tune a custom Gemini model on your own labeled data for a specific task.
- Optional: livestream with Donny Cheung, Scott Coull, Ewa Dominowska, Chris Grier, Anant Nawalgaria, Karthik Raman.
Day 5 — MLOps for Generative AI
- Theory: how to adapt MLOps practices to Generative AI, and how to use the Vertex AI toolset for foundation models/GenAI applications (including AgentOps for agent-style applications). Whitepaper: "MLOps for Generative AI".
- No dedicated codelab — instead, the livestream is a live code walkthrough/demo of the
goo.gle/agent-starter-packrepo (reading the repo beforehand is encouraged). - Optional: livestream with Sokratis Kartakis, Gabriela Hernandez Larios, Ivan Nardini, Anant Nawalgaria, Elia Secchi, Michael Styer, Saurabh Tiwary.
Bonus and After the Course
- Bonus notebook: introduces some additional Gemini API capabilities not covered in the main 5 days (no accompanying whitepaper/podcast).
- After completion, Kaggle suggests continuing with the "AI Agents Intensive" guide for deeper, more hands-on work with AI agents.
Quick Summary by Day
| Day | Topic | Hands-on highlights |
|---|---|---|
| 1 | Foundational LLMs & Prompt Engineering | Gemini API, prompt techniques, evaluation with an autorater |
| 2 | Embeddings & Vector Stores/DB | RAG QA system, text similarity, Keras classifier with embeddings |
| 3 | Generative AI Agents | Function calling + SQL, order-taking agent with LangGraph |
| 4 | Domain-Specific LLMs (SecLM, Med-PaLM...) | Google Search grounding + visualization, fine-tuning a custom Gemini model |
| 5 | MLOps for GenAI | Vertex AI tooling, agent-starter-pack, AgentOps |
See also: LLM Learning Resources · LLM Overview · Agents Overview