natharmatron/MediSight.AI
💻🔒 A local-first full-stack app to analyze medical PDFs with an AI model (Apollo2-2B), ensuring privacy & patient-friendly insights — no external APIs or cloud involved.
What's novel
💻🔒 A local-first full-stack app to analyze medical PDFs with an AI model (Apollo2-2B), ensuring privacy & patient-friendly insights — no external APIs or cloud involved.
Code Analysis
6 files read · 3 roundsA local FastAPI application that loads a 2B-parameter LLM (Apollo2-2B) to extract text from medical PDFs and generate patient-friendly summaries, risk assessments, and recommendations via prompt engineering, with a simple chat interface for follow-up questions.
Strengths
The frontend is clean and well-organized with proper error handling and UX states. The backend correctly implements a local-first approach with sensible prompt design for medical report analysis.
Weaknesses
No tests exist, the code hardcodes CUDA device usage, uses a global in-memory dict for context (no multi-user support), and includes a completely unrelated gini.py file. The README has broken URLs throughout and the core 'AI' logic is entirely delegated to a pre-trained model with no custom algorithms.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
5
Contributors
2
Files
8
Active weeks
2
Repository
Language
Python
Stars
2
Forks
0
License
—