RAG & Context Engineering
Building Production-Grade AI Systems. An eight-week, build-first program: you leave with a deployed retrieval-augmented AI system, the evaluation data to defend every design choice, and the judgment to know when each technique earns its place.
How this program works
Every week follows the same professional rhythm: a live session where concepts are taught and the lab is built together, a checkpoint that becomes a permanent part of your capstone system, a weekend challenge that stress-tests what you built, plus curated reading and video. Nothing is throwaway — by Week 8 your weekly checkpoints assemble into one complete, deployed, evaluated RAG system running on a document collection you chose in Week 1.
From day one you work like a working engineer: local LLMs via Ollama, a real git repository, reproducible environments, and the rule that every claim gets tested with code — no copying framework defaults on faith.
The eight weeks
When (and When Not) to Retrieve
What language models actually know, where they hallucinate, and the engineering decision that defines every RAG system.
Classical Retrieval: BM25 and the Inverted Index
Fifty years of search engineering in one week — and the baseline every later technique must beat.
Semantic Retrieval: Embeddings and Vector Search
Meaning as geometry: how embeddings find what keywords miss — and the chunking decision that quietly dominates quality.
Hybrid Retrieval and Reranking
Neither keywords nor vectors win alone. Production systems fuse both — then let a heavier model re-order the shortlist.
Evaluation: Diagnosing Failures Systematically
'It looks right' is not evaluation. This week you build the measurement machine that turns debugging from guesswork into diagnosis.
End-to-End RAG and Multihop Retrieval
Wiring retrieval into generation properly — then handling the questions no single document can answer.
Agentic RAG: Retrieval-Aware Workflows
From pipeline to agent: the model decides when to search, judges its own context, and re-queries until it can answer.
Production: Secure, Observable, Deployable
The final mile: your system meets adversaries, latency budgets, and real users. Capstone week.
The capstone
Your final session is a defense, not a demo. You present your deployed system live, walk through the design decisions behind it — chunking, hybrid fusion, reranking depth, agent-vs-pipeline routing, injection defenses — and back every choice with your own evaluation numbers. You deliver it twice: once for a technical audience, once for stakeholders. That second version is often the harder one, and the more valuable skill.