Turn a folder of raw patient case files into a set of de-identified research records — entirely on your own computer. Your identified files never leave your browser. Only de-identified text is ever produced, so no protected health information is exposed to us, to the platform, or to anyone else.

How a run works

• Point the tool at a folder of case files — one subfolder per patient, and a patient may have as many documents as you like. (A flat folder of single-file patients works too.)
• Each patient’s documents are read and de-identified on your machine using three independent passes: deterministic HIPAA Safe-Harbor pattern matching, an on-device named-entity model, and a structural quasi-identifier screen.
• Confident identifiers — names, dates, record numbers, contact details, locations, ages 90+, and the rest of the Safe Harbor list — are removed automatically. Anything genuinely ambiguous is set aside into a short red-flag review for you to confirm. By design this rarely fires; when it does, nothing is hidden from you.
• Runs are resumable and support accrual — add patients to the same folder later and re-run; already-processed patients are skipped automatically.

What you receive

• A set of de-identified records, one per patient, keyed by neutral identifiers only — your patients’ folder names never appear in the output.
• A private crosswalk that stays on your machine and is never uploaded. There is no path back to your source files from the de-identified data.
• The red-flag review file with your decisions, plus a provenance receipt — the engine version and per-pass results for every record.

Scope and honesty

Target: HIPAA Safe Harbor. This is a de-identification aid — it does not certify de-identification, and you remain the responsible party for the data you release. It de-identifies; it does not analyze. The output is exactly the input the IPAK-EDU Case Series and Case Review reporters accept.