https://github.com/anshulranjan2004/sqlrx-docnexusai
LLM-powered natural language to SQL for healthcare claims search. Supports multi-table joins, filters, and structured results.
https://github.com/anshulranjan2004/sqlrx-docnexusai
agentic-ai crewai langchain-python streamlit text-to-sql
Last synced: about 1 month ago
JSON representation
LLM-powered natural language to SQL for healthcare claims search. Supports multi-table joins, filters, and structured results.
- Host: GitHub
- URL: https://github.com/anshulranjan2004/sqlrx-docnexusai
- Owner: AnshulRanjan2004
- License: mit
- Created: 2025-07-12T19:00:01.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2025-07-28T11:40:51.000Z (11 months ago)
- Last Synced: 2025-07-28T13:27:24.547Z (11 months ago)
- Topics: agentic-ai, crewai, langchain-python, streamlit, text-to-sql
- Language: Python
- Homepage:
- Size: 10.2 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 🩺 SQLRx – DocNexus AI
**Natural Language to SQL for Healthcare Claims Analytics**
LLM-powered natural language to SQL for healthcare claims search. Supports multi-table joins, filters, and structured results. This prototype allows commercial teams to query healthcare claims data using natural language. Designed for scalability and healthcare-specific accuracy, it converts free-form questions into actionable SQL, supports interpretation, compliance checks, and intelligent refinement.
## Architecture Diagram
## 🧠 Architecture Highlights
- **Agentic Workflow**: Built using **LangChain** for modular and scalable task delegation.
- **Model-Agnostic LLM**: Gemini Flash used for dev; swappable with GPT-4 or Claude for better results.
- **Database**: Lightweight **SQLite** backend (easily extendable to Postgres/Snowflake).
- **RAG Support**: Schema ingested via text files or direct introspection to improve SQL generation.
- **Auto Execution & Feedback**: Queries are run and errors/results are fed back for 2–3 retry loops.
## 🧰 Tech Stack
| Component | Tool/Library |
|------------------|------------------------|
| Frontend UI | Streamlit |
| LLM Orchestration | LangChain + Python |
| Language Models | Gemini 2.5 Flash / GPT-4 |
| Backend DB | SQLite (extensible) |
## ✨ Key Features
- 🔍 **Natural Language to SQL**
- 🛡️ **Compliance Checks (HIPAA/PHI)**
- ♻️ **Regenerate Query if Unsatisfied**
- 📊 **Query Accuracy Metrics + Logging**
- 💡 **Schema + Domain-Aware Interpretation**
- ⚙️ **Direct DB Execution from UI**
## 📝 Notes & Assumptions
- Domain concepts (ICD-10, CPT, HCP, etc.) are embedded into all prompt workflows
- Schema is interpreted dynamically to adapt to structural changes
- Prototype is privacy-aware with simplified PHI/PII validation
- System is optimized for dev use but ready for enterprise scale-up
## 📸 Screenshots
> See the `Documentation and Screenshots.pdf` for visuals of query generation, execution, compliance validation, and logging UI.
## 📂 Folder Structure
```
├── src/
│ ├── langchain_agents.py # Modular agent definitions
│ ├── streamlit_app.py # UI logic
│ └── utils/ # Helpers, schema loaders
├── config/ # Crew AI prototype (discarded)
├── vanna-ai/ # Vanna AI prototype (discarded)
├── SchemaNotes.txt # Optional schema hints
└── README.md
```