CoolFace
Apppublic

deepakvish001/sql-repair-env-new

sourceHugging Faceupdated 5mo agoView on Hugging Face
0likes
App README

SQL Query Repair Lab 🔧

An OpenEnv-compliant reinforcement learning environment where AI agents learn to identify and fix broken SQL queries — with a full interactive dashboard for human users too.

🚀 Features (8 Advanced)

1. 🎨 Interactive SQL Editor

  • —macOS-style code editor with syntax-aware textarea
  • —Load original broken query, clear, character/line count
  • —Keyboard shortcut: Ctrl+Enter to submit

2. 📋 Task Dashboard & Sidebar

  • —3 tasks (Easy/Medium/Hard) with color-coded difficulty badges
  • —Real-time progress bars showing best scores per task
  • —One-click task switching with visual active state

3. 🚀 Real-time Query Grading

  • —Submit fixes and get instant scoring with partial credit
  • —Success / Partial / Fail states with animated result panels
  • —Progressive hints revealed after failed attempts

4. 📊 Score Breakdown Visualization

  • —Per-criterion partial scores (syntax, columns, row count, ordering, etc.)
  • —Color-coded +/- indicators for each grading criterion
  • —Clear feedback on exactly what's right and what needs fixing

5. 📋 Step History Timeline

  • —Visual timeline with green/yellow/red dot indicators
  • —Score, feedback message, and submitted query for each step
  • —SVG score progression chart across attempts

6. 🗃️ Database Schema Explorer

  • —Interactive table view with columns, types, constraints, notes
  • —Sample data preview for each table
  • —Parsed directly from DDL schema definitions

7. 🔍 Diff Viewer

  • —Side-by-side comparison: broken query vs. your fix
  • —Unified diff view with color-coded additions/removals
  • —Toggleable live diff in workspace

8. ▶️ Query Runner & Analytics

  • —Execute any SELECT query against the database (no scoring)
  • —Validate SQL syntax without submitting
  • —Performance analytics: average score, tasks completed, best score
  • —Session timeline tracking all attempts

HTTP API

MethodEndpointDescription
GET/Frontend dashboard
POST/resetStart episode: {"task_id": "..."}
POST/stepSubmit fix: {"fixed_query": "..."}
GET/stateFull internal state
GET/tasksList all tasks
GET/healthLiveness probe
GET/schema/{id}Parsed schema for a task
POST/validateValidate SQL syntax: {"query": "..."}
POST/executeExecute query (read-only): {"query": "..."}
POST/hintGet hint: {"task_id": "...", "step": 0}
GET/analyticsSession analytics summary
POST/analytics/resetReset analytics data

Setup & Usage

Local (Python)

bash
pip install -r requirements.txt

# Start the server
uvicorn main:app --host 0.0.0.0 --port 7860

# Open browser to http://localhost:7860

Docker

bash
docker build -t sql-repair-lab .
docker run -p 7860:7860 sql-repair-lab

# Open browser to http://localhost:7860

Run Baseline Agent

bash
export HF_TOKEN=your_hf_token
export MODEL_NAME=meta-llama/Llama-3.3-70B-Instruct
export API_BASE_URL=https://router.huggingface.co/v1
python inference.py

Project Structure

sql-repair-lab/
├── Dockerfile
├── README.md
├── openenv.yaml
├── requirements.txt
├── pyproject.toml
├── app.py               ← HF Spaces entry point
├── main.py              ← FastAPI server (enhanced with 6 new endpoints)
├── inference.py          ← Baseline RL agent
├── static/
│   └── index.html        ← Full frontend dashboard (single-file, no build step)
├── server/
│   ├── __init__.py
│   └── app.py
└── env/
    ├── __init__.py
    ├── models.py          ← Pydantic models
    ├── database.py        ← SQLite setup + seed data
    ├── tasks.py           ← Task definitions + graders
    └── environment.py     ← Core environment (reset/step/state)

Tasks

#TaskDifficultyBugsExpected Baseline
1Fix Syntax ErrorsEasy1~0.85
2Fix Logic ErrorsMedium2~0.55
3Fix Complex QueryHard3~0.40

License

MIT