Cognit Care
Clinical diagnostic copilot deployed on private infrastructure
Cognit Care is an LLM-powered diagnostic assistant that surfaces differential diagnoses, missed-finding alerts and guideline citations directly inside the clinician's EHR — without a single byte of patient data leaving the hospital network.
The challenge
Clinicians wanted LLM support but compliance and patient-safety teams had blocked every cloud-based offering. The bar was clear: zero data egress and zero hallucinated guidance.
Our approach
We fine-tuned a Gemma 4 model on a curated, de-identified clinical corpus, then deployed it on the hospital's on-prem GPU cluster behind a FHIR-aware retrieval layer. Every suggestion is grounded in a cited guideline; ungrounded outputs are suppressed.
The outcome
Validated against 12,000 clinical cases, Cognit Care improved differential coverage by 31% and cut documentation time by 22% — without a single data-protection incident.
What we built
- Custom Gemma 4 clinical fine-tune on de-identified data
- FHIR-aware retrieval grounded in cited guidelines
- On-prem GPU deployment with zero data egress
- Hallucination suppression with refusal-by-default behaviour
Vector-01
Autonomous coding fleet shipping PRs across 40+ repositories