# TriageAI — Primary Care Pre-Visit Intake Summarizer (MVP)

TriageAI is a lightweight AI-powered prototype that summarizes patient-reported pre-visit intake information into concise, clinician-friendly documentation for primary care visits.

The goal is to reduce time spent on repetitive intake review while preserving clinical judgment, safety, and human oversight.

---

## Problem

In primary care, clinicians often spend valuable visit time reviewing basic intake information that could have been collected and summarized beforehand. This leads to:
- Reduced face-to-face time for complex issues
- Repetitive questioning
- Documentation burden
- Lower patient and clinician satisfaction

---

## Solution

TriageAI allows patients to complete a structured intake form prior to their appointment. An AI model then:
- Synthesizes the information into neutral, clinician-ready summaries
- Highlights relevant medical history and social factors
- Flags missing or unclear information for follow-up
- Avoids diagnoses, recommendations, or clinical decision-making

The output is designed to **support — not replace — clinician judgment**.

---

## What the AI Does (and Does Not Do)

### ✅ The AI:
- Summarizes patient-reported information
- Uses conservative, clinician-appropriate language
- Flags missing or unclear details
- Produces structured, predictable output
- Maintains explicit safety boundaries

### ❌ The AI does NOT:
- Provide medical advice
- Make diagnoses
- Suggest treatments
- Score risk or triage urgency
- Replace clinical decision-making

---

## Example Output

> *“The patient is a 25-year-old female presenting for obesity management. She has a history of diabetes and hypertension. The onset of obesity was noted 3 years ago, with a reported worsening trend. Current medication includes Ozempic. The patient reports daily alcohol intake and a history of vaping. No allergies were reported.”*

---

## Intended Users

- Primary care clinicians
- Clinical operations teams
- Product managers exploring AI workflows in healthcare
- Interviewers evaluating applied AI judgment in regulated domains

---

## Design Principles

- **Safety-first:** conservative language, no clinical decisions
- **Transparency:** all outputs are clearly labeled as AI-generated
- **Clinician control:** AI assists documentation, not care decisions
- **Scope discipline:** focused on intake summarization only

---

## Tech Stack

- **Frontend:** Streamlit
- **AI Model:** OpenAI (structured JSON outputs)
- **Language:** Python
- **Deployment:** Streamlit Community Cloud

---

## Data & Privacy Notes

- This prototype is intended for **demonstration with synthetic or example data only**
- Do **not** enter real patient identifiers or protected health information (PHI)
- No data persistence or clinical system integration is implemented

---

## Disclaimer

This project is an educational prototype only.  
It is **not intended for clinical use**, diagnosis, or treatment.  
All outputs are AI-generated from patient-reported information and have not been verified by a clinician.

---

## Project Status

This project is intentionally scoped as a **demonstration MVP** and is considered feature-complete for its purpose.  
Further work (e.g., clinician evaluation, workflow integration, governance review) would be required for real-world deployment but is out of scope for this repository.