Projects / Flagship
EDScribe
An AI workspace for emergency physicians — build, save, and run reusable clinical prompt templates.
What it is
EDScribe is a physician-facing AI workspace built for emergency medicine. Rather than a single fixed assistant, it lets clinicians create their own library of reusable prompt templates — each tuned for a specific workflow such as drafting a procedure note, summarizing a patient handoff, generating a differential, or refining clinical communication.
Every template runs inside a dedicated chat workspace where physicians can attach files, hold parallel conversations, swap models or providers mid-session, and refine the underlying system prompt on the fly — saving improvements back to the template for next time.
Core capabilities
Template management
Create, edit, version, and organize reusable AI prompts with names, descriptions, and system prompts. Templates are archived rather than hard-deleted to preserve a clinical-use audit trail.
Multi-thread workspaces
Each template launches into its own chat workspace with multiple independent conversation threads — useful for exploring variations of the same documentation task in parallel.
File attachments
Upload clinical notes, images, lab reports, or reference documents directly into a chat for the AI to reference when generating drafts or summaries.
Provider & model switching
Swap AI providers and models from inside the workspace without recreating templates — useful for benchmarking output quality across the rapidly evolving model landscape.
In-session prompt refinement
Adjust the system prompt mid-conversation to steer output, then save the refined version back to the template for future use.
Database-backed audit log
Every template create, update, deletion, and chat action is recorded in PostgreSQL for compliance review and longitudinal analysis of how AI is being used in clinical workflow.
Technology
EDScribe is built on a modern, type-safe full-stack: Next.js 15 (App Router) and React 19 on the front end, Prisma ORM against PostgreSQL for persistence, and a UI assembled from Tailwind CSS v4, shadcn/ui components, and Radix UI primitives for accessibility.
Status & roadmap
EDScribe is currently an early-stage MVP under active development. Template CRUD, the chat workspace UI, and audit logging are all working end-to-end and persist to PostgreSQL. The next milestones are wiring real AI provider execution into the chat layer, adding persistent file storage, and standing up authentication so the system can be piloted in a controlled clinical environment.
This is a research project, not a clinical product. EDScribe is not currently approved for use in patient care.