Text2SQL with User Cost Tracking

An AI-powered multi-agent Text-to-SQL and Business Intelligence platform that converts natural language into verified SQLite queries, executes them safely, and generates dynamic visualizations with per-user cost tracking.

---

⚡ Features

• 🤖 AI Query Studio: Natural language to SQL with automatic schema understanding & BI analysis.
• 🔄 Self-Healing Workflow: LangGraph multi-agent loop with automatic SQL validation and up to 3 repair attempts.
• 📊 Robust Visualizations: Programmatic profiling for Bar, Line, Pie, Scatter, KPI, Table, and No-Results views.
• 🗄️ Isolated Datasets: Upload custom .csv, .xlsx, .json datasets isolated per user.
• 💻 SQL Console & Schema Explorer: Direct SQL query execution & interactive schema inspection.
• 💰 Token & Cost Tracking: Per-user tracking of tokens, estimated API costs, latency, and query history.
• 🔭 Observability: Integrated Langfuse tracing for agent calls, retries, and token metrics.

---

🏗️ Multi-Agent Workflow

User Question → Schema Analysis → SQL Generation → Validation
                                                      │
   ┌──────────────────────────────────────────────────┴──────────────────────────────────────────────────┐
(Valid)                                                                                              (Invalid)
   ↓                                                                                                     ↓
Execution → BI Data Profiling → Chart Normalization → Visualization                         Self-Repair Node (Up to 3 Retries)

---

📊 Supported Visualizations

Chart Type	Best Used For
Bar Chart	Categorical comparisons (vertical columns & horizontal bar layouts)
Line Chart	Chronological trends or ordered numerical sequences
Pie Chart	Proportions of a whole (up to 6 slices + automatic "Other" grouping)
Scatter Plot	Dual numeric metric correlation analysis
KPI Card	Single key aggregate values (currency, percentage, units, compact notation)
Table View	High-cardinality, multi-dimensional, or text-heavy query results
No-Results	Clear empty state when 0 rows match criteria

---

🛠️ Tech Stack

• Frontend: React + Vite, Tailwind CSS, Recharts, Lucide Icons
• Backend: Python, FastAPI, SQLite, Pandas, LangGraph, Groq API
• Observability & Auth: Langfuse, Google OAuth 2.0

---

🚀 Quickstart Guide

1. Setup Backend

cd backend
python -m venv venv
# Windows: .\venv\Scripts\activate | Linux/macOS: source venv/bin/activate
pip install -r requirements.txt

Create backend/.env:

GROQ_API_KEY=your_groq_api_key

GOOGLE_CLIENT_ID=your_google_client_id
JWT_SECRET=your_jwt_secret

# Optional: Langfuse tracing
LANGFUSE_SECRET_KEY=
LANGFUSE_PUBLIC_KEY=
LANGFUSE_HOST=https://cloud.langfuse.com

Run Backend:

uvicorn server:app --reload

---

2. Setup Frontend

cd frontend
npm install

Create frontend/.env:

VITE_GOOGLE_CLIENT_ID=your_google_client_id
VITE_API_URL=http://127.0.0.1:8000
VITE_LANGFUSE_URL=https://us.cloud.langfuse.com

Run Frontend:

npm run dev

App will be running at http://localhost:5173.