tekvo
← All case studies
Healthcare · 2025

Retrieval platform over 2.1M clinical documents

Hybrid search and citation-first retrieval, built so every generated answer points back to the source paragraph a clinician can verify.

94%
Citation accuracy
1.1s
P95 retrieval
2.1M
Docs indexed

What was the problem?

Clinicians would not trust an answer they could not verify. A previous vector-only pilot returned plausible summaries with no traceable source and was shelved.

How did Tekvo build it?

  1. 01

    Rebuilt retrieval as hybrid: BM25 for exact clinical terminology, dense vectors for paraphrase, reciprocal-rank fusion on top.

  2. 02

    Chunked on document structure rather than token count, so a citation resolves to a paragraph a clinician can read.

  3. 03

    Scored citation accuracy as a first-class eval metric, gated in CI alongside answer quality.

What changed?

94% citation accuracy against a clinician-labelled golden set, 1.1s P95 across 2.1M documents, and a review workflow that finally passed clinical governance.

More systems in production

All case studies ↗
Start a scoping sprint

Still running the business on software from the dinosaur age?

Two-week scoping sprint, fixed fee, priced for a mid-size company. You keep the architecture either way.

Let's talk