Skip to content
Case studies

Cutting policy lookup time by 73% at a regional bank

How we built a grounded RAG copilot for compliance and operations teams that handles 14,000 internal queries per month.

Industry
Financial Services
Client
Top-3 regional bank, Southeast Asia
Duration
12 weeks
Stack
Next.jsPostgreSQL + pgvectorAnthropic ClaudeAWS PrivateLinkTerraform

Placeholder case study. Replace with the real engagement details.

Problem

Compliance officers and operations leads spent up to 90 minutes a day looking up internal policies. Policy artifacts lived across SharePoint, Confluence, and a stack of legacy PDF binders. Existing search returned hundreds of unranked results.

Approach

We ran a two-week diagnostic to map question patterns and pick a baseline eval set, then shipped iteratively:

  1. Document ingestion service with deterministic, paragraph-level chunking
  2. Hybrid retrieval (BM25 + dense vectors) tuned with the bank's eval set
  3. A citation-first answer UI with feedback collection
  4. A weekly eval gate so model and prompt updates couldn't regress quality

Outcome

  • 73% reduction in average policy lookup time
  • 98% citation precision on the locked eval set
  • Adoption: 1,400 monthly active users across 4 business units

Ready to ship something real?

Book a 30-minute call. We'll review your roadmap, recommend the smallest useful next step, and tell you honestly whether we're the right partner.

Book a discovery call Email us