CasePrep
Surgical case prep built from your own operative notes.
CasePrep is a web app that prepares a surgeon for an upcoming operation by drawing on their own library of past operative notes rather than generic references. It retrieves the most relevant prior cases for a given attending and procedure, then generates a focused, personalized briefing alongside images pulled from those notes. The result is preparation grounded in how this surgeon and this team actually operate.
The problem
Generic textbooks and surgical atlases do not capture how a specific attending runs a specific operation, or the technical details a trainee logged the last time they scrubbed a similar case. That institutional and personal knowledge sits buried across hundreds of dictated operative notes, effectively unsearchable the night before a case. CasePrep closes that gap by turning a surgeon's own accumulated case record into a queryable preparation tool.
What I built
A standalone web application that ingests a private corpus of several hundred operative notes and a few dozen associated intra-operative images. Given an attending surname plus a procedure or keyword, it assembles a ranked knowledge base of the most relevant prior cases and produces a structured pre-op briefing, complete with a 'from your notes' image strip and a browsable note viewer with lightbox. A companion capture workflow lets the surgeon record a case via a quick note, a post-case voice dictation, and optional teaching audio; dictations are transcribed with a medical-vocabulary speech model and synthesized into a structured case note plus a running per-attending file.
How it works
Retrieval is lexical and metadata-driven — matching on attending and procedure to pull a top-N set of notes into a compact context window — deliberately avoiding embeddings so the system stays correct as the corpus grows across a residency. Inference runs against a local headless model using the surgeon's existing subscription rather than a metered API, with notes read fresh from disk so briefings always reflect the latest record. Voice capture uses on-device speech-to-text with a domain-specific vocabulary prompt, and a clip-review loop cuts short audio snippets around any garbled passages for fast manual correction. New case notes are written back into a version-controlled, sandboxed store.
Where it stands
In active personal use since returning to clinical duty in mid-2026, with the first real cases captured and briefed through the live system. It runs as a persistent always-on service that survives reboots, decoupled from an earlier iOS prototype to avoid app-distribution friction. A key practical finding shaped the design: post-case dictation transcribes cleanly while ambient operating-room audio largely does not, so live teaching capture is now scoped to a close mic. Deferred extensions include feeding images to a vision model, thumbnailing, and multi-user support.