FileGate — Naming Compliance Auditor + AI Intake

Business impact

Turned an unenforced records policy into a measurable data-governance program. FileGate gives leadership an auditable compliance baseline across about 67,000 client files, targets the E&O exposure of misfiled records, and automates document intake with LLM field extraction while keeping humans in the loop for anything uncertain.

Problem

Client records on the shared file server had drifted from the agency's naming and folder policy, which hurt retrieval and created compliance and E&O risk. No one could see where violations were or who owned them.

Solution

Designed, built and put into production a two-part system. The Naming Compliance Auditor encodes the data policy as a machine-readable ruleset, attributes file ownership, locks a baseline compliance score per employee and supports reversible renames. FileGate Intake is an automated document-ingestion pipeline: staff upload a file and an on-site agent uses LLM extraction (Claude API) to identify the client, document type and address, renames it to policy and files it.

Results & highlights

• Treated policy as code (versioned JSON ruleset); anything that can't be derived goes to a human review queue instead of being guessed
• Ran a root cause analysis on stalled weekly scans: throughput fell from about 178 items/s on the office LAN to about 7 over the VPN. Fixed it with a network gate, incremental commits and explicit failure states, then migrated the job to an always-on host
• Production scan: 67,240 items processed in 24 minutes with 0 errors (2026-09-22)
• Human-in-the-loop design: unmatched documents are parked for review, never mis-filed
• Quality gate: 62 automated tests passing; a 388,000-row database migration verified before go-live
• Drove adoption with change-management deliverables: a 33-slide training deck, two narrated videos, a 24-question assessment and two SOPs (2026-09-24)

Competencies demonstrated

• Data governance & compliance: Encoded the agency's records and naming policy as a versioned, machine-readable ruleset with per-employee compliance baselines and reversible renames.
• AI / LLM integration: Automated document intake with Claude API field extraction (client, document type, address), with a human review queue for anything unmatched.
• Root cause analysis & performance tuning: Diagnosed a roughly 25× throughput gap (LAN vs VPN) and re-architected the job with a network gate, incremental commits and explicit failure states.
• Software quality: 62 automated tests passing and a 388,000-row database migration verified before go-live.
• Change management & training: Drove adoption with a 33-slide training deck, two narrated videos, a 24-question assessment and two SOPs.

Architecture

Notes

• Completed: in production since 2026-09-23. Compliance-rate improvement will be published once rename cycles begin.

Open-source companion

naming-policy-validator on GitHub: Policy-as-code for shared drives: audits every folder and file name against a records policy written in JSON and scores compliance with a reason code per violation.

Related writing

• Why a weekly file scan stalled over the VPN, and how the fix brought it to 24 minutes
• Windows silently drops trailing periods from folder names, and your naming policy should know