Build a RAG system.
Understand every layer of it.
This isn't a workshop where you copy code and leave with a chatbot. You'll build a production-style RAG system from scratch — document loading to LLM integration — and understand why every piece exists.
₹599 → ₹499 with code FLAT100
A chatbot that can answer questions about your own documents.
The chatbot itself isn't the point — it's the medium. Every one of these pieces gets built by hand, not imported as a black box.
Document loading
Getting real files into a shape a model can use.
Chunking
Why you can't just embed a whole PDF.
Embeddings
Turning meaning into numbers.
Vector databases
Storing meaning so it can be searched.
Similarity search
Finding what's actually relevant.
Retrieval
Deciding what the model gets to see.
Prompt engineering
Shaping how the model uses context.
LLM integration
Connecting retrieval to generation.
Seven missions. No finished project handed to you.
You clone a starter repository — not a finished one. Each mission is a set of TODOs, and each one opens with a discussion before any code gets written.
- 01
Teach the AI to read documents
Load raw files and turn them into usable text.
- 02
Split documents into meaningful chunks
Break text down without breaking its meaning.
- 03
Generate embeddings
Convert chunks into vectors a machine can compare.
- 04
Store vectors in a database
Give your embeddings a home you can query.
- 05
Retrieve relevant context
Pull back only what's relevant to a question.
- 06
Connect an LLM
Turn retrieved context into a real answer.
- 07
Improve answer quality
Diagnose bad answers and fix the pipeline, not just the prompt.
Not “use this library.”
Why does this library need to exist?
Every component is introduced by answering why it exists before writing code. You won't memorize an API — you'll understand the engineering decision behind it.
Why do we chunk documents?
Why can't we embed the whole PDF?
What actually is an embedding?
Why do vector databases exist?
Why does retrieval happen before generation?
Why does the model hallucinate?
A system you understand is one you've watched fail.
Tiny chunk sizes. No overlap. Wrong top-K values. Poor prompts. Missing retrieval entirely. You'll break the pipeline on purpose, watch the answers get worse, and learn to diagnose why — instead of blindly trusting a system that happens to work.
The same architecture powers a lot more than a Q&A bot.
You'll leave having built the foundation behind:
Nida Botawala
Builder and workshop host focused on making AI engineering concepts click through hands-on, first-principles teaching — no black boxes, no hand-waving.
with code FLAT100
- Live, online workshop
- Full workshop recording included
- Starter repo with 7 guided missions
- Certificate of completion
- Post-workshop resource pack
August 9, 2026 · 2:00–6:00 PM IST