Healthcare access remains unequal globally. Many patients delay care because they don't know if symptoms warrant a doctor visit, what medications are safe, or where to find care. We were inspired to build an AI system that acts as a 24/7 health triage assistant—reducing unnecessary ER visits while catching emergencies before they escalate.
What it does
MediFlow AI is a multi-agent healthcare platform that:
- Analyzes symptoms with clinical accuracy using NLP
- Recommends safe OTC medications while checking allergies and drug interactions
- Monitors condition progression and detects worsening symptoms
- Escalates emergencies instantly when red flags appear
- Connects patients to nearby care (hospitals, specialists, pharmacies)
- Schedules follow-ups and recommends pre-visit diagnostic tests
- Tracks recovery over time with transparent clinical reasoning
It's a care navigator, not a doctor replacement—designed to empower informed decisions and guide patients to appropriate care.
How we built it
- 14 specialized agents (LangGraph/CrewAI) handling intake, diagnosis, medication safety, monitoring, escalation, care recommendations, and follow-up
- Medical NLP (BioBERT, MedSpaCy) for symptom extraction and normalization
- FastAPI backend for agent orchestration and real-time decision-making
- PostgreSQL + ChromaDB for patient history and medical knowledge retrieval
- Google Maps API for nearby healthcare facility recommendations
- React/Next.js frontend for patient-friendly symptom input and care guidance
Challenges we ran into
- Medical accuracy without overconfidence: Balancing helpful suggestions with clear disclaimers that AI isn't a diagnosis
- Symptom ambiguity: Same symptoms can indicate vastly different conditions—we solved this with probabilistic confidence scoring and escalation thresholds
- Real-time drug interaction checking: Built a safety validation layer across all medication recommendations
- Emergency detection logic: Defining red-flag thresholds that catch genuine emergencies without false alarms
- Explainability in healthcare: Users need to understand why the system made a recommendation—implemented transparent reasoning in every agent
Accomplishments we're proud of
- Zero unsafe recommendations: Drug interaction + allergy checking prevents harmful suggestions
- Longitudinal monitoring: Tracks symptom trajectories over days, not just snapshots
- Smart escalation: Automatically routes high-risk patients to emergency care
- Explainable AI: Every recommendation includes clinical reasoning patients can understand
- End-to-end workflow: From symptom input → diagnosis estimation → medication safety → doctor summary (complete care journey)
What we learned
- Healthcare AI requires multiple specialized agents, not one generic model
- Transparency builds trust—in healthcare, users must understand the reasoning
- Safety is non-negotiable—one missed drug interaction undermines the entire system
- Longitudinal data matters—single-visit analysis misses critical patterns
- Care continuity is key—follow-ups and monitoring reduce readmission rates
What's next for MediFlow AI
- Clinical validation: Partner with hospitals for real patient data testing and refinement
- Telemedicine integration: Direct video consultation links to verified doctors
- Multilingual support: Localize for regional languages (Hindi, Telugu, etc.)
- Wearable integration: Connect to health trackers for continuous vital monitoring
- Predictive analytics: ML models to forecast disease progression before symptoms worsen
- Insurance + pharmacy partnerships: Real-time cost and coverage information for medications
Log in or sign up for Devpost to join the conversation.