CHP Report Automation

The Problem

PI law firms need CHP traffic collision reports to build cases. Getting them means navigating a clunky government portal, calling offices to get missing data fields (crash time, officer ID, NCIC codes), and waiting days. Firms were losing time on every single case.

The Approach

The pipeline: email intake monitors requests from law firms, AI figures out what data is missing, voice AI calls the right CHP office (out of 101 statewide) to fill gaps, browser automation submits the portal form, report gets delivered. I wrote specs for each piece and directed coding agents to build it.

Key Design Decisions

1. 1.Data provenance: law firm data is tagged as unverified, CHP officer data as authoritative. The system never mixes them up. This was a spec requirement from day one.
2. 2.Cumulative data: every interaction carries all confirmed data forward. Nothing gets lost between calls. CHP officers don't want to repeat themselves, so the system never asks twice.
3. 3.Office rotation: never call the same CHP office twice in one day. Learned this from building the system and testing against real offices.
4. 4.Escalation: any confusion routes to a human supervisor. The system never lies about being AI. Non-negotiable.

Results

Stack