Case studies
Four anonymized engagements that show how the service lines work in practice — what the problem was, what was built, and why it stuck.
Local-LLM Assistant
An in-app AI assistant over sensitive work-log data, running entirely on a locally-hosted model — selected by a scripted evaluation loop and shipped behind three layers of anti-fabrication guardrails.
Read the case study →Email Security Harness
An email-processing dashboard feeds third-party email into an LLM — so a deterministic, un-promptable security layer screens every message before it can ever reach the model.
Read the case study →GIS Platform CI Security
A GIS web platform shipped code, dependencies and infrastructure completely unscanned — the fix was a repeatable, fully offline security stage that runs on every build without any external service.
Read the case study →This Site's Chat Assistant
The assistant answering questions on this site is itself a live case study — provider-agnostic architecture, a deterministic pre-prompt security gate, layered cost guardrails, and a golden-set evaluation before any change ships.
Read the case study →Facing a similar problem?
Book a call