Defend

40% better retrieval accuracy

It’s your most-quoted number, so it’s usually the first one tested.

On your resume

Improved enterprise retrieval accuracy by 40% through document chunking, embeddings, semantic search, grounding, and response-quality optimization.

Cross-questions

4
A strong answer covers
  • Name the metric: recall@k, hit rate, MRR, nDCG or answer accuracy.
  • Give the baseline and the result, and say whether 40% is relative (0.55 → 0.77) or absolute.
  • Eval set size, how it was built (SME-labeled or synthetic), and that it stayed fixed before and after.
Avoid

Not knowing whether the gain is relative or absolute.

A strong answer covers
  • Walk through the ablation: chunking, embedding model, hybrid search, reranker, query rewriting, metadata filters.
  • Attribute the gain step by step.
Avoid

Crediting every change equally.

A strong answer covers
  • Better retrieval doesn’t guarantee better answers.
  • Measure faithfulness and answer correctness too, with an SME review sample.
Avoid

Treating retrieval accuracy as answer quality.

A strong answer covers
  • A golden set in CI, monitoring as documents change, and a re-indexing plan for embedding upgrades.
Avoid

A one-time improvement with no guardrail.

Your numbers

Interviewers probe numbers first. Write exact values so you never have to improvise.

Next claim