Legallens AI: Medical Lens (V2)

Legallens AI-Medical-Lens is a fine-tuned version of Meta-Llama-3.1-8B, optimized to act as a "Street-Smart" translator for complex medical documents. It simplifies technical jargon to a 5th-grade reading level and provides a color-coded risk assessment for patients.

Model Description

Project: Part of the Legallens Hackathon suite.
Task: Medical Text Simplification & Risk Categorization.
Base Model: Meta-Llama-3.1-8B (4-bit Quantized via Unsloth).
Training Data: medical_v2_balanced.jsonl (700+ curated medical samples).
Output Format: Structured bullet points including Explanation, Risk Level, and Reason.

Risk Categorization System

The model uses a specific logic to flag documents for users:

🟢 Green (Informational): Standard health facts, routine medications, and general terminology.
🟡 Yellow (Standard Warning): High-risk procedures (surgery), emergency symptoms, or safety warnings.
🟡 Yellow (Financial Caution): Insurance traps, out-of-network costs, and predatory billing clauses.

Training Details

Optimization: Fine-tuned using Unsloth for 2x faster training and 70% less memory usage.
Hardware: Trained on a single NVIDIA T4 GPU (16GB VRAM).
Technique: LoRA (Low-Rank Adaptation) fine-tuning.
Loss: Final training loss achieved: ~0.82 - 0.88.
Balancing: V2 utilizes oversampling of "Yellow" warning classes to prevent label bias.

Intended Use & Limitations

Intended Use: This model is designed to help patients understand the "gist" of their medical paperwork by simplifying complex terminology and identifying potential red flags or high-cost items.
Limitations: This is an AI tool, not a medical professional. It may occasionally "over-sensitize" (flagging minor health issues or standard procedures as 🟡 Yellow). Always consult a qualified medical professional for actual health advice and clinical decisions.