learn.ai

Week 2 of 5

RAG and context engineering

Ground the assistant in your documents with hybrid retrieval, citations and access control, and keep its context small.

0%

Pick a start date to see real dates.

0%
Today

Build this week

  • Ingestion pipeline (parse, chunk, embed, store in pgvector)
  • Hybrid search (BM25 + vectors) and a reranker, measuring whether each one helps
  • Incremental updates keyed by chunk hash
  • Role metadata filtered at query time (access control)
  • Citations that link to the source chunks
  • Summarising long conversations and truncating large tool outputs
  1. Day 81 h 30 min

    Retrieval that works

    Notes
  2. Day 91 h 30 min

    Context as a budget

    Notes
  3. Day 101 h 30 min

    Parse and chunk

    • 1 h 30 min
    Notes
  4. Day 111 h 30 min

    Embed and store

    • pgvectorOllama
      1 h 30 min
    Notes
  5. Day 124 h 45 min · weekend

    Hybrid search

    • PostgreSQL: Full Text Search
      1 h 30 min
    • Contextual retrieval
      2 h 30 min
    • 45 min
    Notes
  6. Day 133 h 45 min · weekend

    Trust and limits

    • 1 h
    • 1 h 15 min
    • Effective context engineering for AI agents
      1 h 30 min
    Notes
  7. Day 141 h 20 min

    Week review

    • 50 min
    • 30 min
    Notes

If you have time

  • 45 min

Checkpoint

Answer out loud, then rate yourself.

What chunking strategies do you know, and how do you pick one?
Why does RAG fail?
When is hybrid search better than pure vector search?