💫 Lunara — AI-Powered Women's Health Triage Assistant
Rutgers HealthHack 2025 — Team 16
Inspiration
Every day, women’s health call centers face overwhelming complexity — agents must sort through mountains of patient histories, symptoms, and sub-specialty decisions. Misrouted referrals delay care, especially for high-risk pregnancies and cancer patients.
We asked: What if triage could be smarter, faster, and more accurate — from the very first call?
What It Does
Lunara is an AI-powered triage assistant built specifically for women’s health.
It empowers call center agents to make evidence-based referral decisions in under 30 seconds.
Agents enter only minimal information (name, DOB, initial symptoms).
Lunara then:
- Uses Natural Language Processing (NLP) to interpret patient-reported symptoms
- References prior medical history for contextual accuracy
- Generates a ranked list of conditions, sub-specialties, and suitable doctors
- Displays confidence scores, top recommendations, and a summary card for transparency
The result: faster, smarter, and more accurate triage — routing patients to the right provider the first time.
How We Built It
Tech Stack Overview
- Frontend: React.js
- Backend: FastAPI
- Machine Learning: TF-IDF + Logistic Regression for symptom → condition → specialty mapping
- Database: AWS Aurora Serverless (PostgreSQL)
- Infrastructure: AWS (EC2 + RDS + S3)
Pipeline
- Agents input minimal patient information via the frontend — first name, last name, date of birth, and an initial symptom or concern.
- The backend processes the input and performs NLP-based symptom interpretation, referencing prior medical history and context.
- The ML model ranks the most probable conditions and sub-specialties, generating confidence scores for each.
- The backend then returns a clarifying question to the agent — dynamically selected from the model’s uncertainty space — which can be answered with Yes / No / Skip.
- Each response instantly updates the confidence scores and ranked suggestions, refining the triage in real time until the model reaches a high-confidence recommendation.
- The final output includes:
- Top-ranked sub-specialty and doctor recommendation
- Updated confidence scores with supporting explanations
- A summary card for the call agent
- Top-ranked sub-specialty and doctor recommendation
- Every triage interaction (including clarifying questions, responses, and final decisions) is logged with the algorithm version, timestamp, and confidence metrics, ensuring full auditability, transparency, and HIPAA compliance.
Admin Tools
- Modify or override mappings between symptoms and conditions
- Review logs and monitor model performance across 140 tracked conditions
- Continuously improve model accuracy without downtime
🧠 Model Development
We trained a TF-IDF Logistic Regression model using anonymized hospital call transcripts.
- Monte Carlo permutation tests validated that performance was above random chance.
- Achieved 93% accuracy in routing to the correct specialty.
- Tuned for high recall to minimize missed high-risk cases.
The model merges clinical reasoning with administrative efficiency, adapting continuously to real-world hospital data.
🌎 Impact
- 25% of women’s health calls are misrouted today — Lunara reduces this drastically.
- Cuts average triage time from minutes to under 30 seconds.
- Reduces cost of care and wait times for patients.
- Improves safety by ensuring high-risk cases reach the right specialists immediately.
Lunara is more than a hackathon project — it’s a blueprint for equitable, data-driven triage in women’s healthcare.
🏆 Accomplishments We’re Proud Of
- Deployed a working AI triage system from scratch in under 48 hours
- Achieved 93% routing accuracy with transparent confidence metrics
- Created a fully functional React dashboard for agents and admins
- Built real-time AWS integration with auditable triage records
- Collaborated across engineering, data science, and clinical domains
🔮 What’s Next for Lunara
- Integration with live hospital call center systems
- Expansion to other specialties (e.g., pediatrics, oncology)
- Real-time speech-to-text triage for phone call automation
- Continuous learning with live feedback loops
- Implementing a per-user dashboard with authentication using AWS Cognito
- Fully deploying the app using AWS Lambda and S3 + Cloudfront
👩💻 Team 16
A cross-disciplinary group of student engineers and student doctors from:
- New Jersey
- Illinois
- Maryland
Working together to make women’s healthcare faster, smarter, and safer.
🌐 Official Links
Log in or sign up for Devpost to join the conversation.