AI / COMPUTING · λ 590 NM · PROJECT
Elymetis
On-prem AI that turns scattered clinical records into one auditable case model.
Problem
A patient's record is not one record. It's a structured database export, a stack of scanned PDFs, and a large volume of clinical free text — discharge summaries, referral letters, notes — each written for a different reader and none of it joinable. Anything needing a complete view of a case starts with a human reassembling it by hand.
Two constraints shape the problem: the data cannot leave the building, and the output has to be checkable. A mapping you can't audit is worse than no mapping — it produces confident answers with no way to verify where they came from.
Approach
All three input types resolve into a single unified case model, on-premise.
Lineage is a first-class output: every field carries where it came from — which document, which extraction, which transformation — so any value can be traced back to the source line that produced it. Data-quality checks run inside the pipeline rather than as a downstream report, so conflicts, gaps and low-confidence extractions surface at mapping time instead of being discovered later by whoever trusted the output.
On-prem by design. No data egress, which is a hard requirement in European clinical settings and rules out a lot of otherwise convenient architecture.
Context
Built for a project with EPA CC.
Result
evaluation figures not yet supplied.