RAG chatbot walkthrough: from PDFs to answers with citations
The full pipeline in one sitting: chunking, embeddings, pgvector and a grounded prompt.
Build with LLM APIs.
Recording in production
The full written walkthrough is below. Get an email when the video goes live.
Chapters and walkthrough
- 00:00
Architecture
Ingest, chunk, embed, retrieve, generate, and where each can go wrong.
- 06:30
Ingestion and chunking
Extract PDF text and split on headings with a small overlap.
- 14:00
Vector search
Store embeddings in pgvector and query the nearest chunks.
- 22:10
Grounded answers
A prompt that cites sources and admits when it doesn't know.
- 29:00
Evaluating
A small question set to measure changes rather than guess.
More videos
AI at Work.
ChatGPT in 10 minutes: everything a beginner needs
Accounts, your first prompts, custom instructions and the mistakes to avoid.
AI for Coding.
Ship your first feature with Claude Code
Plan, implement, test and commit a real change from the terminal.
Automation & Agents.
Build an AI agent in n8n that uses your tools
An agent that can look up customers in a sheet and draft replies, step by step.