somAI-frontend / README.md
arshenoy's picture
Update README.md
487fa84 verified

A newer version of the Streamlit SDK is available: 1.56.0

Upgrade
metadata
title: SomAI | Sovereign Medical Intelligence
emoji: 🧬
colorFrom: indigo
colorTo: green
sdk: streamlit
sdk_version: 1.35.0
app_file: app.py
pinned: true
license: apache-2.0
short_description: Privacy-first Medical AI running on Llama-3-OpenBioLLM.

SomAI: Sovereign Patient Intelligence

SomAI is a privacy-first, local-execution medical AI designed to bridge the gap between clinical jargon and patient understanding. Unlike wrapper-based chatbots that send sensitive data to external APIs (OpenAI/Google), SomAI runs entirely on Sovereign Infrastructure, ensuring HIPAA compliance by design.

Technical Architecture (The "Flex")

SomAI does not rely on rented intelligence. It runs on a quantized version of Llama-3-OpenBioLLM-8B, a State-of-the-Art (SOTA) model fine-tuned on high-quality biomedical data.

  • Model Core: Llama-3-OpenBioLLM-8B (Quantized GGUF)
  • Performance: Achieves ~86% on MedQA (USMLE), outperforming GPT-3.5 and Meditron-70B.
  • Infrastructure: Runs on CPU (16GB RAM container) via llama.cpp.
  • Latency: Optimized for CPU inference using 4-bit quantization (Q4_K_M).

Why "Sovereign" AI?

In the medical domain, data privacy is non-negotiable.

  • Zero Data Leakage: Patient data never leaves this server. It is processed in RAM and discarded.
  • No API Rate Limits: Unlimited inference without per-token costs.
  • Resilience: Can run offline on hospital edge servers.

Capabilities

  1. Clinical Translation: Converts complex diagnosis reports into 5th-grade metaphors.
  2. Mental Empowerment: Integrates psychological support with medical advice.
  3. Safety Rails: Hard-coded refusal mechanisms for prescription/dosage requests.

⚠️ Disclaimer

SomAI is a prototype for educational and informational purposes only. It is not a doctor. It does not provide medical diagnosis, treatment, or prescriptions. Always consult a qualified healthcare provider for medical concerns.


Built for CellStrat × BMCRI Docathon 2025.