Skip to main content
AboutProblemsGuidelinesTerms & conditions

AjraSakha: AI-Powered Agricultural Advisory Platform

1. POP Processing Pipeline​

Project Name: Agricultural Package of Practices (POP) PDF Processing Pipeline

Problem Statement​

565 POP PDF documents across 26 Indian states, in English and 22 Indic languages, are scattered, unstructured, and partially garbled (283 PDFs, 50%). Farmers and advisors cannot access or search this knowledge efficiently.

Solution Overview​

Automated multi-phase pipeline to classify, extract, translate, and convert PDFs into structured JSON/Markdown for RAG-powered AjraSakha chatbot.

Work Done So Far​

  • Phase 1 & 2 complete: 215 PDFs processed (English & Indic).

  • OCR, translation, Docling extraction implemented.

  • Garbled text (Phase 6) preparation: character encoding detection, text reconstruction, font normalization.

Key Features​

  • Multilingual PDF extraction (English + 22 Indic languages).

  • OCR with quality validation.

  • Automated JSON/Markdown output for chatbot ingestion.

Technologies Used​

  • Extraction/OCR: Docling, PyMuPDF, Tesseract OCR, Google Cloud Vision

  • Translation: Google Translate API, IndicTrans2

  • Orchestration: Python CLI pipeline for automated processing

Future Scope​

  • Garbled text recovery for remaining 283 PDFs.

  • Automated Q&A generation.

  • Semantic search across all 565 PDFs.

  • Integration with state agricultural portals.

2. MCP Servers for AjraSakha​

Project Name: AjraSakha MCP (Model Context Protocol) Server Suite

Problem Statement​

Monolithic backend cannot efficiently integrate FAQs, golden datasets, market data, or POP documents. Challenges include scalability, security, maintainability, and performance.

Solution Overview​

Modular MCP server suite built on FastMCP framework, enabling specialized capabilities for chatbot integration.

MCP Servers Implemented​

  • FAQ MCP Server: Semantic search over agricultural FAQs.

  • Golden Dataset MCP Server: Crop & state-specific Q&A retrieval.

  • Market Data MCP Server: Real-time commodity prices from eNAM.

  • Shared Functions & Constants: Utilities, configuration.

Work Done So Far​

  • 4 production MCP servers deployed.

  • Vector search integrated with MongoDB Atlas (BAAI/bge-large-en embeddings).

  • Dockerized, CI/CD pipeline, systemd service management.

Key Features​

  • Modular, secure, scalable backend.

  • Semantic search with vector embeddings.

  • Real-time market data integration.

  • Standardized MCP protocol for easy expansion.

Technologies Used​

  • Backend: Python, FastAPI, FastMCP

  • Database: MongoDB Atlas + Vector Search

  • Infrastructure: Docker, Docker Compose, Systemd

  • LLMs: deepseek-r1:70b, qwen3:1.7b, Osmosis

Future Scope​

  • Caching and query optimization.

  • Horizontal scaling & database sharding.

  • Real-time updates, WebSocket integration.

  • Additional MCP servers for weather, schemes, analytics.

Impact​

  • Democratizes agricultural knowledge across languages.

  • Enables scalable, modular chatbot architecture.

  • Lays foundation for RAG-powered, AI-assisted advisory.

  • Reduces time to access verified agricultural guidance.

  • Facilitates rapid feature development and platform scalability.