Protocolos AI

Inspiration

Clinical professionals spend significant time reviewing patient information and comparing it with complex clinical protocol requirements. Relevant data may be distributed across diagnoses, medications, laboratory results, medical history, and clinical notes.

I created Protocolos AI to automate the initial review while keeping every final decision in the hands of the physician.

What it does

Protocolos AI is a web application that helps physicians:

• Manage patients and their clinical information.
• Load and prepare clinical protocols.
• Compare patients with protocol inclusion and exclusion criteria.
• Detect missing or outdated information.
• Generate an explained result for each patient and protocol.
• Record the protocol version, sources, criteria, and reasoning used.

The application does not replace medical judgment. It provides decision support, traceability, and an organized initial review.

During OpenAI Build Week, I am adding an AI-generated patient briefing that summarizes the most relevant information before a consultation, including diagnoses, treatments, allergies, recent laboratory data, missing information, and points that require medical review.

How we built it

The application uses:

• React and TypeScript for the frontend.
• Supabase and PostgreSQL for authentication and data storage.
• Supabase Edge Functions for backend processes.
• Cloudflare Pages for deployment.
• OpenAI Codex to analyze the existing codebase, implement the new Build Week functionality, review changes, and assist with testing.
• GPT-5.6 to generate structured and traceable patient briefings from clinical information.

The application uses fictional patient data for the competition demonstration.

Challenges we ran into

The main challenge was processing complex clinical information without presenting the AI result as a medical diagnosis.

Other important challenges included:

• Preserving traceability for every result.
• Clearly identifying missing patient information.
• Separating data between authenticated medical users.
• Working

Built With

• cloudflare
• gpt-5.6
• openai-codex
• postgresql
• react
• supabase
• supabase-edge-functions
• typescript
• vite