We hold roughly 25,000 insurance policies as PDFs and scans. Staff currently search them by hand. I need a pipeline that reads them (OCR where needed), indexes them with sensible chunking, answers questions with page-level citations, and says so when it cannot find an answer. Access must follow our existing user permissions.

Looking up one clause or figure can take half an hour.

Key Deliverables:

• System Analysis & Architecture: Reviewing existing system, defining integration points, and confirming approach.

• Implementation & Integration Testing: Building core automation with confidence thresholds and fallback logic.

• Production Deployment & Documentation: Final deployment with monitoring, usage documentation, and cost projections.

Requirements:

• Demonstrated experience with Python, LangChain, Vector Databases, OCR

• Async communication preferred with milestone check-ins.

• Existing user-facing interface must remain unchanged.

• Must include confidence threshold fallback to human review.

Ready to hire as soon as we review qualified candidates. Please share your approach and availability.