A Unified National Marine Data Backbone Fusing INCOIS Physical Oceanography with CMLRE Living Resources
VARUNA is an enterprise-grade agentic AI platform that bridges India's two siloed marine data domains:
- INCOIS (Hyderabad) — Real-time physical and chemical sensor profiles (temperature, salinity, oxygen, chlorophyll) from 3,800+ ARGO floats.
- CMLRE (Kochi) — Marine living resources taxonomy, species distributions, and ecosystem biodiversity records.
VARUNA lets oceanographers, disaster cells, and fisheries authorities ask compound natural-language questions across both domains — and proactively warns them before marine heatwaves or hypoxic anomalies trigger ecological damage.
┌───────────────────────────────────────┬─────────────────────────────────────────────────────────────────────────────┐
│ Capability │ Innovation in VARUNA │
├───────────────────────────────────────┼─────────────────────────────────────────────────────────────────────────────┤
│ 🧠 Multi-Agent Task DAG Planner │ Decomposes compound multi-hop queries into asynchronous dependency DAGs. │
│ 🚨 Proactive Anomaly Agent │ Continuous 6-hr scanner detecting Marine Heatwaves (Hobday 2016) & Hypoxia. │
│ 🐟 CMLRE Darwin Core Fusion │ Spatio-temporal joins (≤50km, ≤7 days) between float data & marine species. │
│ ⚡ Zero-Local-LLM Architecture │ Zero GPU baggage; high-speed async OpenRouter NVIDIA Nemotron-550B. │
│ 🛡️ Strict Zero-Hallucination │ Every numerical value is strictly traced back to a verified SQL row. │
│ 🛰️ Military-Grade Command Center │ Liquid glass dark HUD, Deck.gl WebGL maps, 15+ oceanographic Plotly charts. │
└───────────────────────────────────────┴─────────────────────────────────────────────────────────────────────────────┘
graph TB
UserQuery([User Natural Language Query]) --> APIGateway[FastAPI Gateway /api/v1/agent/chat]
APIGateway --> PlannerAgent[Planner Agent: NVIDIA Nemotron-Ultra 550B]
PlannerAgent --> TaskDAG{ExecutionPlan DAG}
TaskDAG -->|NL->SQL Task| SQLAgent[SQL-Gen Sub-Agent + PostGIS]
TaskDAG -->|Semantic Task| VectorAgent[Hybrid Retrieval Sub-Agent BM25 + Qdrant]
TaskDAG -->|Biodiversity Task| BioAgent[CMLRE Darwin Core Entity Resolver]
SQLAgent --> DB[(PostgreSQL + PostGIS marine_data)]
VectorAgent --> QdrantDB[(Qdrant Vector DB argo_knowledge)]
BioAgent --> BioDB[(PostgreSQL marine_biodiversity)]
SQLAgent --> Synthesizer[Synthesizer Agent: Grounded Provenance]
VectorAgent --> Synthesizer
BioAgent --> Synthesizer
Synthesizer --> CommandCenter[Command Center UI: Live Agent Graph + Plotly]
cd backend
python -m venv venv
.\venv\Scripts\Activate.ps1 # Windows (or source venv/bin/activate on Linux/Mac)
pip install -r requirements.txt
# Configure backend/.env (copy from .env.example with your OpenRouter API key)
# Seed 500+ Indian Ocean marine species records
python -m src.ingestion.seed_biodiversity
# Launch FastAPI server
uvicorn main:app --reload --port 8000- API Documentation:
http://localhost:8000/docs
cd frontend
npm install
npm run dev- Command Center UI:
http://localhost:3000
| Member | Name | Role | Specification |
|---|---|---|---|
| M1 | Aryan Lomte (Lead) | AI Systems Architect & Full RAG Lead | Aryan_Lomte.md |
| M2 | Aditya Yadav | Data Engineer & Backend Lead | Aditya_Yadav.md |
| M3 | Sahil Shah | Predictive ML & Sensor QC Lead | Sahil_Shah.md |
| M4 | Advay Chavan | Frontend Full-Stack Lead | Advay_Chavan.md |
| M5 | Netal Gupta | Geospatial Visualization Specialist | Netal_Gupta.md |
| M6 | Kanishka Sahal | Marine Analytics & Presentation Lead | Kanishka_Sahal.md |
- 📖 Master Handbook: VARUNA.md
- 🏛️ Architecture Overview
- 🤖 Multi-Agent Task DAG Orchestration
- 📊 NetCDF HPC Extraction & Ingestion Pipeline
- 🐟 CMLRE Darwin Core Biodiversity Fusion
- 🚨 Proactive Anomaly & MHW Early-Warning Engine
- 🗄️ Database & Qdrant Vector Schema
- 🧠 OpenRouter Nemotron-550B Cognitive Layer
- 🖥️ Frontend Operations Center UI Specification
- 🛠️ Zero-Docker Local Setup Guide
- 📋 API Contracts & Schema Specifications