pal-palak/RAG-AgroSight
Domain:
agriculturenatural language processingRecord type:
software
Creator:
pal
Host:
Production-grade Agentic Agricultural RAG system built with FastAPI, LangGraph, Mistral AI, Qdrant, and BGE-M3 embeddings. Features hybrid retrieval, multi-tool reasoning, multilingual support, real-time weather & mandi APIs, streaming responses, evaluation framework, caching, and scalable production architecture.
# 🌾 AgroSight RAG System
**Production-grade agricultural Retrieval-Augmented Generation API** serving Indian farmers with crop advisory, disease diagnosis, mandi prices, scheme information, and fertiliser calculations — in English, Hindi, and Gujarati.
---
## Architecture Overview
```
┌─────────────────────────────────────────────────────────────────────┐
│ AgroSight RAG Pipeline │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ Data Sources │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │113 PDFs │ │ Mandi CSV│ │ Scheme │ │ Disease │ │20K Images│ │
│ │(1 GB) │ │ USDA/FAO │ │ FAQs JSON│ │ KB JSON │ │(Vision) │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│ │ │ │ │ │ │
│ ▼ ▼ ▼ ▼ ▼ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ Hybrid Auto-Select Chunker (chunker.py) │ │
│ │ Semantic | Section | Q&A | Table | Record | Sliding Window │ │
│ └─────────────────────────┬───────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ BAAI/bge-m3 Embedder (1024d, multilingual) │ │
│ │ Dense vectors + BM25 sparse weights (hybrid) │ │
│ └─────────────────────────┬───────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────── …