Focus 01 — Document Intelligence at Scale
Can you build the pipeline that turns COIs, dec pages, endorsements, loss runs, SOVs, and medical records into governed data my underwriters and adjusters can actually trust?
Why this matters
More than 80% of the information that drives underwriting, claims, and servicing lives in COIs, dec pages, endorsements, loss runs, SOVs, and medical records. Owning this means deploying extraction pipelines with ACORD-aligned schemas, full lineage from raw page to decision, and E&O-grade accuracy thresholds. The work is measurable on cycle time, straight-through processing rate, and re-keying labor eliminated.
What I’ve shipped
- unifi.ai COI Compliance Scanner — production AI compliance scanner. Upload a Certificate of Insurance, LlamaParse + Claude Vision extract structured fields, deterministic Python rules engine returns a coverage verdict. Live with Free / $49 / $199 plans.
- DocuScan AI — Enterprise Document Intelligence — full-stack document intelligence platform replacing the paper-to-system pipeline for insurance agencies. 5-factor confidence scoring, 8-stage processing pipeline, multi-org RBAC, BigQuery integration with 323K+ policies. Live at docuscan.5gvector.com.
- 5gVector Endorsement Analyzer — 17 ISO forms, 16 rules, E&O scorer, pdfplumber + Claude extraction. 4 API routers, 5 frontend pages, 81 tests.
- Six-layer architecture — Intake / Understanding / Decision / Action / Truth / Guardrail. Used in production at SWIA and reusable across document types. See Focus 02 for the architecture detail.
- Enterprise Data & AI Transformation at SWIA — full case study including the BigQuery integration of 323K+ policies behind DocuScan.
- ACORD Student Challenge — runner-up. Recognition from the standards body whose forms power the entire industry.
What’s coming next
| Quarter |
Deliverable |
| Q2 2026 |
COI scanner public beta + accuracy scorecard v1 |
| Q3 2026 |
Loss runs as a new document type · Open benchmark suite on GitHub |
| Q3 2026 |
MCP server for COI parsing — submitted to Anthropic directory |
| Q4 2026 |
SOVs as a new document type · Medical-record extraction pilot |
Market signal I track
Hyperscience · Indico Data · Roots Automation · Send · Convr · Cytora · Federato · Ushur · Sixfold · ACORD Forms Working Group · IVANS releases · NAIC SERFF · Applied Epic, Vertafore AMS360/Sagitta, EZLynx, HawkSoft document capabilities · Verisk and CCC Intelligent Solutions earnings on document AI.
Success metrics
- Benchmark accuracy ≥ 97% on critical coverage fields by Q4 2026
- 1,000+ scanner uses by Q3 2026
- Cited as reference architecture in at least one carrier RFP or insurtech blog by Q4 2026