Medi2Mind is a hybrid AI medical assistant that combines multimodal medical image analysis Using Agentic AI with a Mixture of Experts (MoE)-based large language model to simulate real-time consultations across five medical domains: Mental Health, Radiology, General Consultation, Veterinary Medicine, and Medical Imaging.
To build an intelligent, scalable system that:
- Analyzes medical queries and CT/MRI/X-ray images via specialized expert agents.
- Routes queries to domain-specific experts using a dynamic keyword-scoring engine.
- Generates accurate, contextual, and patient-friendly responses.
- Handles high concurrency and cross-domain healthcare scenarios efficiently.
- Languages & Frameworks: Python, PyTorch, Transformers, CUDA
- UI/Frontend: Gradio
- Vision/Image Analysis: OpenCV, PIL
- Agentic Capabilities: Agno Agents (File Upload, Vision, OCR, Prompt Enrichment, LLM Reasoning)
- Deployment: Google Cloud Platform
- Model Backbone:
deepseek-ai/deepseek-llm-7b-base
| Module | Routing Accuracy | Domain Knowledge Accuracy | Response Relevance | Diagnostic Accuracy | Inference Time (s) |
|---|---|---|---|---|---|
| Mental Health | 94% | 91% | 87% | – | 1.21 |
| Radiology | 89% | 94% | 92% | – | 1.35 |
| General Consultation | 85% | 88% | 90% | – | 1.18 |
| Veterinary Medicine | 91% | 89% | 85% | – | 1.25 |
| Image Analysis | 92% | 95% | 89% | 95% | 2.34 |
System handled 100 concurrent queries with 97% success rate.
- Designed a modular expert-routing engine with 91% average accuracy using keyword probability scoring.
- Implemented Agentic AI pipeline to interpret CT/MRI/X-ray scans with 95% diagnostic precision.
- Delivered real-time consultation and second-opinion capabilities for both patients and professionals.
- Validated system across unit, integration, and system testing with high performance under load.
Medi2Mind aims to democratize access to reliable medical advice using scalable, AI-powered healthcare solutions.