Arc 07 · Senior AI Engineer
Retrieval and Knowledge Systems
Build retrieval as a measurable information system: representations, indexing, chunking, filtering, ranking, grounding, and freshness.
Exit capability: Diagnose a RAG system by stage and improve the right component.
- Mapped lessons
- 7
- Published now
- 7
- Full-arc estimate
- ≈17 hours
- Last edited
- 2026-09-20
Lesson sequence
Live units open into complete labs. Planned units stay visible to show the dependency path, but intentionally have no detail route. A withdrawn unit is retained only at its previously advertised URL and is not presented as live.
- 01
The Embedding Contract
LiveVersion task, model, preprocessing, distance, and dimensionality as one inseparable interface.
Systems labIntermediate90 min estimateArtifact: Versioned embedding contract
- 02
Approximate Nearest Neighbors Under a Latency Budget
LiveCompare graph and partitioned indexes through recall, build cost, memory, filtering, and updates.
Build labAdvanced115 min estimateArtifact: ANN index decision record
- 03
Chunking Is a Recall Policy
LiveConnect document structure and answer span length to retrieval candidates and context waste.
Failure labIntermediate95 min estimateArtifact: Chunking coverage audit
- 04
Hybrid Retrieval and Score Fusion
LiveCombine lexical and semantic evidence without comparing incompatible raw scores.
Build labAdvanced110 min estimateArtifact: Hybrid retrieval fusion contract
- 05
Reranking and Context Assembly
LiveSeparate candidate generation, expensive relevance judgment, diversity, and final context packing.
Systems labAdvanced110 min estimateArtifact: Reranking and context assembly plan
- 06
Diagnose RAG by Stage
LiveMeasure corpus, retrieval, reranking, context, generation, citation, latency, and cost separately so failures remain attributable.
Failure labAdvanced125 min estimateArtifact: Stage-isolated RAG evaluation harness
- 07
Retrieval Security, Deletion, and Freshness
LivePreserve tenant ACLs, deletion semantics, provenance, and index freshness through ingestion and ranking.
Design reviewAdvanced105 min estimateArtifact: Retrieval governance audit