← Capabilities
Retrieval-augmented generation (RAG)
not yet earnedI can design a retrieval system, measure retrieval separately from generation, and name the corpus properties that decide whether embeddings, lexical search or hybrid wins.
What this claim costs
- 6 assessed items at tier 2 or above0 of 6 — 6 more to do, from 24 available
- A public artifact built for this claim — retrieval-benchNot built yet
- A documented negative result — where this stops workingNothing recorded. A capability with no known failure boundary has not been explored.
- Defended aloud, against questions you did not write1 available, none done
- Recalled cold, a week later, without notes6 scheduled, none done
The proof
retrieval-bench
Two corpora chosen because they fail differently — dense proper-noun rate documents against near-synonymous prose — with the crossover point where hybrid stops paying for itself.
What it has to contain
- Two corpora chosen because they fail differently, with the properties stated
- Retrieval measured separately from generation
- Lexical, embedding and hybrid compared on both corpora
- The crossover point where hybrid stops paying for itself
Where it stops working. Not established yet. Anyone can ship a demo; the boundary is the part that is worth reading, so this claim is not earned until it is written down.
Skills underneath
- Embedding retrieval
- Chunking strategy
- Hybrid and lexical search
- Reranking
- Evaluating retrieval
- Access-controlled retrieval
- Ingestion and freshness
24 assessed items across the specified courses evidence these, 0 done.