0/35FDE path lessons understood. Go to your dashboard.

 

Topic 02 of 10

Retrieval

Answers from the client's own documents: parsing, chunking, search, reranking, permissions, and measuring whether any of it works.

  • 11 lessons
  • 6.8 h reading
  • 12.8 h hands-on
  • 3 on the FDE path
Start lesson 01
Progress0 / 11

Lessons

  1. 01Parsing and chunking documents that fight backPath · stage 3 LabTables, headers and scans survive ingestion, so retrieval has something real to find. Pro
  2. 02Picking embeddings on your corpus, not a leaderboardSpecialisationEmbeddings cap retrieval quality; measure them on the customer's own queries. Pro
  3. 03Vector stores and HNSW tuningSize, tune and price the index against the real corpus and its recall target. Pro
  4. 04Hybrid search and rerankingPath · stage 3Keywords catch the codes and IDs that vectors miss; a reranker puts the best first. Pro
  5. 05Rewriting, routing and fusing queriesHandle the vague and multi-hop questions people actually type. Pro
  6. 06Retrieval that plans, checks and retriesWhen retrieval becomes an agent, and what each extra loop costs. Pro
  7. 07Retrieval that respects who can see whatPath · stage 3Filter by the user's document permissions before anything reaches the model. Pro
  8. 08Scans, forms and OCR at scaleMost enterprise input is scanned paper; extract it accurately and affordably. Pro
  9. 09When questions need a graphSpecialisationRelationship questions break chunk stores; know when a knowledge graph pays off. Pro
  10. 10Keeping the knowledge base fresh LabSync, re-index and catch staleness before answers quietly go out of date. Pro
  11. 11Contract review and legal workSpecialisationReview contracts against a playbook, cite every clause, and never invent law. Pro