| I first posted about PrivateScribe.ai ~1yr ago and have recently jumped back intent on bringing it to a functionality that makes it actually usable by non-technical users. One year ago it worked but only the bare minimum. Since then I've gotten ⭐️74 github stars!⭐️ and have had a few meetings with people that has inspired me to push it forward. PrivateScribe is a fully local, open source AI transcription platform using FasterWhisper, pyannote, and Ollama, built with Vite/Flask/SQLite. I am an ER physician in my second life and I've approached a lot of this project with a focus on privacy and specifically HIPAA workflow requirements. The medical world has been flooded with dozen(s) of AI-transcription startups focusing on free tiers with the ever-questionable data policies or permanent subscriptions and I'm still strongly of the opinion this is a solvable problem locally especially for small clinics, therapists, and beyond medicine into law, counseling, and personal use. Excited to share the major updates: A signed, notarized, bundled macOS app - launch ETA this Friday! Ollama, pyannote, everything bundled into the application so no separate installs - detects a system Ollama if you've already got one otherwise it handles the setup and model pulls. Onboarding Wizard - walks the first user through the admin setup, hash key storage (and a brief overview for those who've never seen one), ollama set up, selecting use case to pre-populate templates, etc. Speaker diarization - labels who said what and then allows fully customizable editing afterwards as needed. Security First - Everything is local and encrypted - database is encrypted with SQLCipher 256bit encryption, audio files are encrypted (if you choose to save them at all) with 256bit encryption. The application makes zero network calls after the initial install. Admins can rotate keys. Server-side sessions, password hashing, two factor auth, brute force lockouts, role-based access. Audit trail - every user facing action is logged and stored with a hash-chain for verification. Option to use the standard note signatory flow (approve a transcript -> finalize a formatted note -> sign to make immutable -> timed addenda can be then added as needed). Full admin dashboard - user management, role assignments, data retention, everything configurable (that way a personal user doesn't need to be bothered by the HIPAA focused functionality). Everything is under the MIT license. Would love feedback on anything/everything. Github is here [link] [comments] |