A generative AI course where you build, not just prompt

Generative AI, from the inside

Generative AI is the part of AI that creates text, images, code or audio. For an engineer, the useful questions are how these models work, and how to turn them into products that are correct, safe and affordable. The program answers both, in this order:

1. How the models work. You train neural networks in PyTorch and build a working mini-GPT on Indian-language text (Module 5), so attention, tokens and context windows become concrete.
2. Building LLM applications. Structured, validated outputs, cost and latency control, and a real service behind an API (Module 6).
3. Retrieval-augmented generation. Answers grounded in your own documents, with citations and measured retrieval quality (Module 8).
4. Evaluation, guardrails and security. Regression tests for AI behaviour and defences against prompt injection (Module 9).
5. Agents. Tool use, memory and human approval for risky steps, through the Model Context Protocol (Module 10).
6. Fine-tuning and open models. When to fine-tune, how to do it on free GPUs, and how to serve the result (Module 11).

What you can show at the end

Each module ends in a portfolio project: a mini-GPT, an LLM data service, a bilingual RAG assistant, an eval suite, an agent and a fine-tuned model published on Hugging Face. Start with the free first lesson and see how lessons work.