SQL+AI Datathon Submission — ClinIQ

[learnsol]

Project name

ClinIQ

Description

Project: ClinIQ — RAG-Based Medical Chatbot

🏗️ Architecture Summary

• Frontend: Vanilla HTML/CSS/JavaScript
• Backend: Fast API (Python 3.12)
• Database: SQL Server 2025 with native VECTOR columns
• Vector Search: SQL stored procedures using VECTOR_DISTANCE('cosine')
• AI Generation: GitHub Models (GPT-5-mini) with Ollama/Azure OpenAI fallback
• Deployment: Docker Compose (development), Azure Container Apps (production)

✨ Key Features

1. Retrieval-Augmented Generation (RAG)

   • Vector embedding of user queries (text-embedding-3-small)
   • Semantic search on SQL Server VECTOR columns
   • Multi-factor relevance scoring (cosine + recency + citations)

2. Sourced AI Responses

   • Every answer includes citations (PubMed PMIDs, NCT IDs)
   • Evidence retrieved from SQL Server
   • Structured JSON responses with source metadata

3. Multiple AI Backends

   • GitHub Models (default, cloud-based) Or Local Ollama for development
   • Azure OpenAI (production)

4. Medical Data Sources

   • PubMed articles (vectorized abstracts)
   • Clinical trials (characteristics + eligibility)

5. Production-Ready

   • OWASP Top 10 security
   • Rate limiting & input validation
   • Non-root Docker containers

📊 SQL + AI Integration

• Native VECTOR type on medical articles table (1536-dim embeddings)
• Stored procedures for semantic search with structured filters
• Hybrid retrieval: Vector similarity + attribute-based filtering (trial status, patient eligibility)
• Data API Builder provides REST gateway to SQL (no direct SQL from app)

🚀 Quick Start

git clone https://github.com/YOUR-ORG/ClinIQ.git
cd ClinIQ
cp .env.example .env  # Set GITHUB_TOKEN for AI features
docker-compose up --build
# Open http://localhost:8000

See [SETUP.md]for detailed instructions.

Type

Chat App

Project Repository URL

https://github.com/learnsol/ClinIQ

Project video (please verify that the link works)

https://youtu.be/F-EwAHDBmkQ?si=gzdCSr0m7T2KENPV