MediScan AI

Inspiration

Healthcare reports are often complex, filled with medical jargon, and difficult for non-experts to understand. We wanted to build something that bridges this gap, an intelligent assistant that can simplify medical data and provide meaningful insights instantly. The idea was to make health information more accessible, understandable, and actionable for everyone.

What it does

Our project is an AI-powered health risk analyzer that allows users to upload medical reports (PDFs or images). It extracts the text using OCR, processes it with an LLM, and provides:

Simplified explanations of medical terms Risk predictions based on report data Context-aware answers to user questions Multilingual support for accessibility

How we built it

Frontend: React (Vite) for a fast and responsive UI Backend: FastAPI for handling APIs and processing AI Layer: Groq-powered LLM for analysis and explanations OCR Engine: Extracts structured text from PDFs/images Integration: Axios for seamless frontend-backend communication Logic Layer: Hybrid approach combining LLM + rule-based risk detection

We designed the system to be modular so each component (OCR, LLM, API) can evolve independently.

Challenges we ran into

Parsing inconsistent OCR outputs from different file formats Ensuring reliable and safe JSON responses from the LLM Managing API key issues and switching providers (Anthropic → Groq) Aligning frontend-backend communication during rapid development Making the output both medically relevant and easy to understand

Accomplishments that we're proud ofBuilt a fully functional end-to-end AI health analyzer from scratch

Successfully integrated OCR + LLM + rule-based system into one pipeline Enabled users to upload reports and get instant, simplified insights Resolved major integration issues (API alignment, CORS, LLM reliability) Designed a modular system that’s easy to scale and extend

What we learned

Practical integration of LLMs into real-world applications Handling and cleaning unstructured medical data Importance of fallback systems alongside AI for reliability Debugging full-stack issues across frontend and backend Working collaboratively under hackathon pressure

What's next for MediScan AI

Improve medical accuracy with better datasets and validation Add visual dashboards for clearer report insights Expand multilingual and accessibility features Integrate real-time health tracking (wearables, APIs) Enhance risk prediction with more advanced models

Built With

• axios
• fastapi
• groq-llm
• ocr
• react(vite)