OptimizeSeniorAI engineeringAuto-graded
Raise retrieval recall from 60% to 90%
Diagnosing why an AI search misses answers and improving it against a measured target is a senior-level task.A retrieval pipeline over a small product-docs corpus recalls about 60% of the relevant passages. Get it to 90%. Everything is in play — how documents are chunked, which embeddings you use, whether you add a lexical signal or a reranking pass, and whether you rewrite the query before searching. The first thing to build is the measurement, because without an eval harness you cannot tell an improvement from a lucky run.45 min
python
You're done when
- An eval harness exists and reports recall over the labelled corpus.
- Real embeddings are used rather than a placeholder or a random vector.
- The chunking strategy is improved beyond the naive fixed split.
- Retrieval combines signals — a hybrid lexical-plus-vector approach, or a reranking stage.
- Query rewriting is applied so the search is not limited to the user's literal phrasing.
- Measured recall reaches the target, demonstrated by the harness rather than asserted.
What it measures
RAG and retrieval systemsKnowing which levers actually move recall, and in what order they are worth pulling.
AI evaluation and QABuilding the measurement first so each change can be judged against a number rather than an impression.
Iterating on evidenceImproving a pipeline by measuring, changing one thing, and measuring again.
Your environment
Terminal + editorA full shell in the browser
Claude CodePre-authenticated, no key setup
Isolated microVMYours alone, wiped afterwards
Start this lab
45 min
Start labYou'll sign in to Anthropos to start. The lab opens in a private cloud workspace with Claude Code ready to use.AI Labs is part of Anthropos Workforce. Not on your plan yet? Book a demo