An intelligent, AI-powered system designed to automatically analyze Reinforced Cement Concrete (RCC) structural drawings (Foundations, Slabs, and Beams) and verify their compliance with Indian Civil Engineering Standards.
This application leverages a sophisticated Multi-Agent AI Architecture combined with RAG (Retrieval-Augmented Generation) to deeply analyze complex RCC structural engineering drawings.
It specifically checks designs against:
- IS 456:2000 — Indian Standard Code of Practice for Plain and Reinforced Concrete
- SP 34 — Handbook on Concrete Reinforcement and Detailing
By automating the tedious compliance review process, this tool helps structural engineers rapidly identify missing parameters, structural risks, and code violations before construction begins.
- 📄 Multi-Element Support — Effortlessly analyze Foundation, Slab, and Beam structural drawings from a single interface.
- 🤖 Multi-Agent Orchestration — An intelligent Orchestrator Agent automatically classifies the uploaded drawing type and dynamically routes it to the correct AI specialist (Foundation, Slab, or Beam Agent).
- ✅ Interactive Validator Loop — If crucial data is missing from the drawing, the system prompts the user. A dedicated Validator Agent actively assesses user-supplied answers for engineering plausibility before allowing the report generation to proceed.
- 📚 RAG-Enhanced Reports — The final AI report doesn't guess; it uses vector embeddings to retrieve and reference the exact IS code clauses that justify its compliance decisions.
- 💾 Supabase Integration — User accounts, session history, and full reports are securely saved to a PostgreSQL database utilizing Row Level Security (RLS).
- 📊 Professional Exports — View beautifully formatted reports in the interactive dashboard or download them instantly as PDFs for client delivery.
- 🔄 Resume Capability — Started an analysis but didn't finish providing missing data? Seamlessly resume incomplete reports directly from the History dashboard.
- Python 3.11 or higher
- uv (blazing fast Python package manager)
- Node.js & npm (for the React frontend)
- OpenRouter API Key (Provides access to the Gemini 2.5 Flash Vision model)
- Supabase Project (Database and Authentication)
# Navigate to the backend directory
cd server
# Copy environment template
cp .env.example .env
# Edit .env and add your keys:
# OPENROUTER_API_KEY=your_openrouter_api_key_here
# SUPABASE_URL=your_supabase_project_url
# SUPABASE_KEY=your_supabase_service_role_key
# Install all Python dependencies using uv
uv sync
# Initialize and populate the RAG Vector Database (ChromaDB)
uv run python ingest.py
# Start the FastAPI server on port 8000
uv run uvicorn main:app --reload# Open a new terminal and navigate to the frontend directory
cd client
# Copy environment template
cp .env.example .env
# Edit .env and add your safe frontend keys:
# VITE_SUPABASE_URL=your_supabase_project_url
# VITE_SUPABASE_ANON_KEY=your_supabase_anon_key
# VITE_API_BASE_URL=http://localhost:8000
# Install Node dependencies
npm install
# Start the Vite development server
npm run devThe frontend application will now be running at http://localhost:5173.
The system operates using a state-of-the-art multi-agent workflow:
- Orchestrator — The "manager." It quickly analyzes the visual file to determine if it's a foundation, slab, or beam, ensuring the correct domain logic is applied.
- Specialist Extractor — A vision-enabled agent tailored to the specific structural element. It extracts highly specific engineering parameters (e.g., steel percentage, lap lengths, cover depth) and flags any mandatory data missing from the drawing.
- Validator — The "gatekeeper." If the user manually inputs missing data, the Validator ensures the response is an actual engineering value (and not empty or nonsensical) before advancing the pipeline.
- RAG Reporter — The "auditor." It cross-references the extracted parameters against embedded chunks of the actual IS Code from the Vector DB, generating a definitive compliance verdict (Compliant / Non-Compliant / Warning).
├── client/ # React / Tailwind Frontend
│ ├── src/
│ │ ├── api.js # Axios configuration and backend calls
│ │ ├── App.jsx # React Router setup
│ │ ├── components/ # Reusable UI (Sidebar, FileUpload, ReportDisplay)
│ │ ├── context/ # React Context for Supabase Auth state
│ │ └── pages/ # Views (Dashboard, History, Login)
│ └── package.json
│
├── server/ # FastAPI / LangChain Backend
│ ├── main.py # Core API endpoints routes
│ ├── llm_handler.py # Vision AI Specialist agent execution
│ ├── llm_service.py # Orchestrator, Validator, and RAG Reporter logic
│ ├── prompt.py # Central registry for all LLM system prompts
│ ├── vector_db.py # ChromaDB retrieval and querying standard
│ ├── data_loader.py # Utility for chunking IS Code markdown
│ ├── ingest.py # Embedding script for the RAG pipeline
│ ├── auth.py # Supabase JWT authentication middleware
│ ├── database.py # Supabase python client init
│ ├── SP34_md/ # Source folder for Raw IS Code markdown files
│ ├── chroma_db/ # Generated vector embeddings output
│ └── pyproject.toml # uv package dependencies
| Component | Technology Used |
|---|---|
| Frontend UI | React 18, Vite, Tailwind CSS, React Router |
| Backend API | FastAPI, Pydantic, Python 3.11 |
| AI / LLMs | Google Gemini 2.5 Flash (Vision & Text) via OpenRouter |
| Vector Database | ChromaDB (Local high-performance vector store) |
| RAG Embeddings | HuggingFace sentence-transformers/all-MiniLM-L6-v2 |
| Auth & Database | Supabase (PostgreSQL, Row Level Security, JWT) |
| PDF Generation | ReportLab (Python PDF creation) |
| Document Parsing | PyMuPDF (PDF image extraction) |
This project is open-source and available under the MIT License. Contributions are highly encouraged! Feel free to submit a pull request or open an issue if you have suggestions for new features or IS codes to include.