MEDILINK

Inspiration

Overcrowded government hospitals and overworked doctors often face unnecessary patient visits for symptoms that could be treated at home. Many patients also lack medical knowledge to decide whether they should go to the hospital or rest at home. We wanted to build an AI-powered telemedicine assistant that triages patient symptoms and takes intelligent action — reducing load on healthcare systems and saving time for patients.

What it does

Our Telemedicine Assistant is an agentic AI system that guides patients through a digital self-checkup and takes action based on the outcome: Collects patient symptoms via chat or form Passes data to a Granite triage agent trained to classify severity Based on the outcome: Low-risk? Recommends home remedies Medium/high-risk? Auto-books a teleconsultation with an available doctor Sends the user a confirmation message, calendar invite, and pre-consultation checklist

How we built it

Multi-agent framework: Built using CrewAI Granite model: Used [Granite-13B or similar] to act as a clinical triage decision-maker Booking system: Integrated a mock scheduling API or Google Calendar API to auto-book teleconsults Frontend: Simple chat interface using Flask for symptom input and interaction Workflow orchestration: Agent 1: Symptom collector Agent 2: Triage classifier (Granite) Agent 3: Scheduler agent

Challenges we ran into

Getting the triage agent to classify edge cases accurately Integrating real-time scheduling and simulating availability Ensuring privacy when handling sensitive health-related data Designing a clean handoff between multiple agents

Accomplishments that we're proud of

Built an end-to-end autonomous pipeline from symptom intake to teleconsult booking Successfully simulated a Granite-powered triage system Created a modular multi-agent setup that can be expanded to other health services Kept everything simple enough to be usable by low-tech users

What we learned

Built an end-to-end autonomous pipeline from symptom intake to teleconsult booking Successfully simulated a Granite-powered triage system Created a modular multi-agent setup that can be expanded to other health services Kept everything simple enough to be usable by low-tech users

What's next for HACKATHON PROJECT

Integrate with real patient portals or health APIs Add multilingual support for broader accessibility Use fine-tuned models or medically-aligned LLMs (e.g., Med-PaLM) Add voice input/output for low-literacy users Get feedback from healthcare workers to iterate on real-world use

Built With

• crewai
• flask
• google-calendar