⚠️ This repository contains project documentation only. Source code is maintained in a private repository.
An AI-powered environmental compliance assistant that combines RAG (Retrieval-Augmented Generation) with real-time web search to help landowners navigate federal and state environmental regulations.
flowchart TD
User([User]) --> Frontend[Lovable Frontend]
Frontend --> API[FastAPI Backend]
API --> A2[Web Search Agent]
API --> BG[Background Analysis]
BG --> A1[RAG Search Agent]
BG --> A3[Gap Analysis Agent]
BG --> A4[URL Recommendation Agent]
A1 --> ChromaDB[(ChromaDB\nVector Store)]
A2 --> Gemini[Google Gemini\nGrounded Search]
A3 --> Supabase[(Supabase)]
A4 --> Supabase
Worker[Document Worker] --> Supabase
Worker --> ChromaDB
Admin([Admin]) --> Frontend
Frontend --> Upload[Document Upload]
Upload --> Worker
EcoComply AI helps landowners maintain environmental compliance by searching an internal database of 67 federal and state regulatory documents alongside real-time web sources. A multi-agent system performs gap analysis between internal knowledge and current regulations, automatically flagging when the knowledge base needs updating. The admin dashboard surfaces these gaps with priority-ranked URL recommendations for database expansion.
Environmental compliance is a moving target. Regulations change, court decisions alter what's enforceable, and the documents that govern what a landowner can or can't do are scattered across dozens of federal and state agencies.
EcoComply was built for a real estate team that needed to get smarter about environmental compliance — faster than reading every existing regulatory document. But the deeper goal wasn't just to answer questions. It was to build a system that improves as the team uses it.
Every query the system can't answer well becomes a signal. The gap analysis agents automatically identify what's missing from the knowledge base and surface the exact sources needed to fill it. The more the team asks, the more complete the system becomes. The knowledge builds itself.
- FastAPI — REST API backend with async background processing
- ChromaDB — Vector database for document embeddings and semantic search
- LangChain — RAG framework for document ingestion and query synthesis
- OpenRouter — LLM access (Claude 3.5 Sonnet)
- Google Gemini 2.5 Flash — Web search with grounding
- OpenAI text-embedding-3-large — Document embeddings (via OpenRouter)
- Supabase — Database, authentication, and document management
- Lovable — Frontend UI with user query interface and admin dashboard
The /agent endpoint runs two paths in parallel:
- Fast path — Google Gemini 2.5 Flash performs a live web search and returns a response in 2–5 seconds
- Background path — Searches the local ChromaDB vector store (~77k indexed regulatory document chunks), runs a gap analysis comparing RAG vs. web results, and queues recommended documents for admin review if gaps are found
A background worker polls Supabase every 60 seconds for newly approved documents and ingests them into ChromaDB automatically.
Key files:
backend/main.py— FastAPI server & endpointsbackend/simple_agent.py— 4-agent orchestrationbackend/landrag.py— RAG system & vector storebackend/document_worker.py— Background ingestion worker
Sample query and response:
{
"query": "What wetland permits do I need for construction near a stream in Georgia?",
"user_id": "user-123",
"session_id": "session-456"
}{
"status": "success",
"response": "For construction near streams in Georgia, you typically need: (1) A Section 404 permit from the U.S. Army Corps of Engineers under the Clean Water Act for any discharge of dredged or fill material into waters of the U.S., (2) A Section 401 Water Quality Certification from Georgia EPD...",
"sources": [
"33 CFR Part 323 - Permits for Discharges of Dredged or Fill Material",
"Georgia Water Quality Standards (Aug 2022)",
"1987 Corps Wetland Delineation Manual"
],
"response_time_ms": 6200
}Multi-agent gap analysis (admin view):
{
"alert_level": "YELLOW",
"gap_summary": "RAG database covers federal CWA requirements but is missing Georgia's 2024 updated 303(d) list and recent EPD guidance on stream buffer variances.",
"recommended_urls": [
{
"url": "https://epd.georgia.gov/watershed-protection-branch/water-quality-georgia",
"priority": "high",
"reason": "Georgia 2024 impaired waters list not in RAG database"
}
]
}- 76,914 document chunks indexed from 67 federal and state regulatory documents
- 5–10 second response time for user-facing web search queries
- 95% excellent-to-perfect search agent evaluation scores (assessed by Google Gemini 3 Pro)
- 128 test runs completed during development and validation
- 3-tier gap detection (GREEN / YELLOW / RED) automatically identifies knowledge base gaps
Prerequisites
- Python 3.10+
- API keys for: OpenRouter, Mistral, Google (Gemini), and Supabase project credentials
1. Install dependencies
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt2. Configure environment
cp .env.example .envFill in .env with your credentials:
OPENROUTER_API_KEY=...
MISTRAL_API_KEY=...
GOOGLE_API_KEY=...
SUPABASE_URL=...
SUPABASE_KEY=...3. Start the backend
cd backend
uvicorn main:app --reload --host 0.0.0.0 --port 8000You should see:
🚀 Starting Land Compliance RAG API...
[WORKER] 🚀 Document ingestion worker started
INFO: Uvicorn running on http://0.0.0.0:8000
4. Verify it's running
curl http://localhost:8000/health
# {"status": "ok"}5. Send a query
curl -X POST http://localhost:8000/agent \
-H "Content-Type: application/json" \
-d '{
"query": "What wetland permits do I need for construction near a stream?",
"user_id": "user-123",
"session_id": "session-456"
}'- Add support for additional state regulatory databases (currently covers Georgia, Florida, South Carolina)
- Implement scheduled regulatory update monitoring with automated alerts
- Add document versioning to track regulatory changes over time
- Expand OCR capabilities for scanned historical documents
- Build batch query mode for compliance audits across multiple parcels