Production-grade AI system that converts natural language software requirements into complete, executable application configurations.
FlowCompiler behaves like a compiler for software applications: you provide a natural language description of what you want to build, and it generates production-ready UI, API, Database, and Auth schemas — validated, repaired, and simulation-tested.
"Build a CRM with login, contacts, dashboard,
role-based access, payments, and analytics."
↓
┌─────────────────────────────────────────┐
│ 1. Intent Extraction │
│ 2. System Design │
│ 3. Schema Generation (UI/API/DB/Auth) │
│ 4. Cross-Layer Validation │
│ 5. Targeted Repair Engine │
│ 6. Runtime Simulation │
└─────────────────────────────────────────┘
↓
Executable Configuration (JSON)
src/
├── api/ # FastAPI routes
│ └── routes.py
├── agents/ # AI pipeline agents
│ ├── base_agent.py # Abstract base with retry/structured outputs
│ ├── intent_extractor.py # Stage 1: NL → structured intent
│ ├── system_designer.py # Stage 2: Intent → architecture
│ ├── schema_generator.py # Stage 3: Architecture → 4 schemas
│ ├── validator.py # Stage 4: Cross-layer validation
│ ├── repair_agent.py # Stage 5: Targeted repair engine
│ ├── execution_simulator.py # Stage 6: Runtime simulation
│ └── pipeline.py # LangGraph orchestrator
├── schemas/ # Pydantic v2 models
│ ├── intent_schema.py
│ ├── system_design_schema.py
│ ├── ui_schema.py
│ ├── api_schema.py
│ ├── db_schema.py
│ ├── auth_schema.py
│ ├── validation_schema.py
│ └── runtime_schema.py
├── runtime/ # Config → executable code generators
│ ├── ui_runtime.py
│ ├── api_runtime.py
│ └── db_runtime.py
├── evaluation/ # Benchmark framework
│ ├── benchmark.py
│ ├── metrics.py
│ └── dataset.json # 20 benchmark prompts
├── tests/ # Test suite
│ ├── conftest.py
│ ├── test_schemas.py
│ ├── test_validator.py
│ └── test_api.py
├── app.py # FastAPI application
└── config.py # Settings management
- Python 3.12+
- Node.js 20+
- OpenAI API key
- PostgreSQL (optional, for full deployment)
git clone https://github.com/dewanggandhi01/FlowCompiler.git
cd FlowCompiler
# Create virtual environment
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS/Linux
# Install dependencies
pip install -r requirements.txt
# Configure environment
cp .env.example .env
# Edit .env and add your OPENAI_API_KEYuvicorn src.app:app --reload --port 8000cd frontend
npm install
npm run devVisit http://localhost:3000 for the dashboard.
docker-compose up --buildFull pipeline execution from prompt to runtime config.
Request:
{
"prompt": "Build a CRM with login, contacts, dashboard, role-based access, payments, and analytics."
}Response: CompilerOutput with all stage results, schemas, validation, simulation, and metrics.
Validate existing schemas across all four layers.
{
"ui": { ... },
"api": { ... },
"db": { ... },
"auth": { ... }
}Repair failing schemas with targeted fixes.
Run execution simulation on generated schemas.
Run the 20-prompt benchmark suite with metrics.
Health check endpoint.
# Run all tests
pytest src/tests/ -v
# With coverage
pytest src/tests/ --cov=src --cov-report=html -v
# Specific test file
pytest src/tests/test_validator.py -v20 prompts (10 normal + 10 edge cases) testing:
- CRM, ERP, LMS, E-commerce, Booking, etc.
- Ambiguous, contradictory, minimal, overly complex inputs
- Non-English prompts
Metrics tracked: success rate, validation failures, repair count, runtime pass rate, latency, cost.
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, React, TypeScript, Tailwind CSS, ShadCN UI |
| Backend | FastAPI, Python 3.12 |
| AI | LangGraph, LangChain, OpenAI (GPT-4o, Structured Outputs) |
| Database | PostgreSQL |
| Validation | Pydantic v2, JSON Schema |
| Deployment | Docker, Vercel, Render |
MIT