Retrieval-Augmented Generation grounds an LLM in your data: chunk and embed documents into a vector store, retrieve the most relevant passages at query time, and hand them to the model. The result answers from real sources, cites them, and hallucinates far less. The hard part isn't the model — it's the retrieval.
Curriculum
For 1:1, built around your documents and your accuracy target
1 · RAG architecture
The full pipeline end to end, where quality is won and lost, and how to reason about it as a system.
2 · Ingestion & chunking
Chunking strategies that preserve meaning, handling messy docs, and metadata that improves retrieval.
3 · Embeddings
Choosing embedding models, dimensions, cost trade-offs, and when to re-embed.
4 · pgvector & Supabase
Storing and indexing vectors in Postgres, similarity search, and keeping it fast at scale.
5 · Hybrid & re-ranking
Combining semantic and keyword search, top-k tuning, and re-ranking for precision.
6 · Grounded answers
Citation-grounded responses, prompt patterns that stay faithful to sources, and refusal when evidence is thin.
7 · Semantic caching
Cosine-similarity FAQ caches to cut latency and per-query LLM cost.
8 · Retrieval evals
Measuring retrieval precision/recall, catching regressions, and knowing the pipeline is good — not hoping.
9 · Ship & scale
Serving RAG behind FastAPI, cost controls, and integrating it into an agent when you need one.
Formats & Details
1:1 Private Mentoring
Built around your corpus and accuracy goals.
- 6–12 hours across sessions
- Fix your real retrieval problems
- Live review + async follow-up
Team Workshop
Standardise RAG patterns across the team.
- 2–3 days, live online
- Labs + your documents
- Eval harness to keep
Prerequisites
- Comfortable with Python
- Basic SQL helps
- No vector-DB experience needed
Frequently Asked Questions
What is RAG?
RAG grounds an LLM's answers in your own data: chunk and embed documents, retrieve the most relevant passages at query time, and pass them to the model so it answers from real sources and can cite them — cutting hallucination.
Why do most RAG systems underperform?
Usually retrieval, not the model — poor chunking, weak embeddings, no hybrid search, and no evals. We focus on measuring and fixing retrieval quality.
Which vector database do you teach?
Primarily pgvector on Postgres/Supabase — production-friendly and keeps data in one place — with concepts that transfer to any store.
How much does it cost?
Depends on format and duration — book a free intro call for a tailored proposal.
Related Training
Agentic AI Training →
Put RAG to work inside a full agent stack.
LangGraph Training →
Wire retrieval into a stateful agent graph.
MCP Training →
Expose retrieval as a tool via MCP.
AI Architecture Mentoring →
Design the whole system around your data.