Sanjivani 2.0
Production-grade AI Agriculture Platform. Hybrid CNN+LLM architecture.
Inference
<100ms
Accuracy
95%
Languages
3 Native
System Architecture & Overview
Production-grade AI agriculture platform. Hybrid CNN+LLM architecture with edge optimization, multilingual support, and comprehensive testing built for real farmers.
The Hybrid Architecture
Sanjivani 2.0 elevates edge diagnosis to a production-grade AI platform. It utilizes a hybrid CNN + LLM architecture: a local CNN model classifies the crop disease instantly on-device, and a remote Gemini 1.5 LLM generates specialized, multilingual treatment plans and actionable insights for the farmer.
"Achieved sub-100ms inference latency for on-device classifications, supporting real-time feedback even in poor connectivity regions."
System Architecture & Specs
FastAPI Backend
High-throughput asynchronous endpoints handling local CNN inference queries and managing external API pipelines.
Gemini 1.5 Engine
Orchestrates complex context retrieval to return structured remedy recommendations in English, Hindi, and Marathi.
Next.js Frontend
Responsive dashboard featuring real-time diagnostic reporting, disease map visualization, and offline state recovery.
Testing & Hardening
To prepare for real-world field conditions and rigorous technical interviews, we established a comprehensive testing pipeline featuring end-to-end integration tests, model accuracy validation, and automated edge-case scenarios.
Technology Stack
- Next.js 14
- FastAPI
- TensorFlow
- Firebase
- Gemini 1.5
- MobileNet
System Highlights
Inference: <100ms. Built with industrial resilience and edge optimization protocol.