A county government: AI automation that never lets resident data leave the building
County government. Engagement: ongoing automation partner since 2025. Related services: AI Workflow Integration, Systems & Data Integration.
A county government wanted the benefits of AI-assisted email handling but could not let resident data reach an external model. We built a triage pipeline that anonymizes personal information before any model sees a message, added retrieval over county records, and built document redaction tooling that quotes a legal justification for every redaction. All of it runs on infrastructure the county controls.
The situation
Staff handled a steady stream of inbound email that needed reading, classifying, and routing, much of it containing residents' personal information. The county also needed to redact council documents before publication, and the applicable legislation requires a stated legal basis for every redaction. Off-the-shelf AI tools failed the data-protection test before the conversation started.
What we built
- AI-assisted email triage with a PII anonymization step in front of the model, so the model classifies and routes without ever seeing a resident's name or details
- Retrieval over county records, so staff can ask questions of their own documents and get sourced answers
- Document redaction tooling designed around the legal requirement to justify every redaction, with a staff review loop before anything is published
- A SQL-to-NocoDB sync that gives staff a usable interface over an internal database
- A major-version upgrade of the county's self-hosted n8n platform, with every workflow tested before and after
The constraints we designed around
- No resident data to any external AI provider: anonymize first, and keep a self-hosted model option available
- Everything deployed on the county's own infrastructure, reached over a private network
- Every redaction traceable to a quoted legal basis, with staff able to accept or override
The outcome
The county has a working pattern for using AI on sensitive data: strip the personal information first, keep the model on infrastructure you control, and keep a person in the loop. The relationship is ongoing, with new projects added as needs change.
Why it matters
This is the engagement that separates us from agencies that wrap a public chatbot in a workflow. AI automation and strict data protection are compatible, and this is what it looks like.