Production-Grade AI System

Sanjivani 2.0

The AI Crop Doctor. Hybrid CNN + GenAI architecture that understands your crops, speaks your language, and works offline.

MobileNetV2
Gemini 1.5

Hybrid Intelligence

Dual-Mode AI

Deterministic vision for 95% accuracy in <100ms, combined with Generative AI for contextual advice and native language explanations.

< 100ms
Edge Inference Latency
3 Langs
English / Hindi / Marathi
Next.js 14
FastAPI
TensorFlow
Firebase
Cloudflare
Render

The 2.0 Shift

From a prototype to a production-grade platform.

Sanjivani 1.0

  • Single Monolith (Flask)
  • Manual Deployment
  • English Only
  • Disease Detection Only

Sanjivani 2.0

  • Microservices (Next.js + FastAPI)
  • CI/CD on Cloudflare & Render
  • Native Multilingual Support
  • Full Farm Management Suite

Engineered for
Resilience.

We moved beyond simple classification. Sanjivani 2.0 implements a **Hybrid AI Pipeline**. The MobileNetV2 model runs first for speed and safety. Only if a disease is detected, the context (weather, location, crop stage) is injected into **Gemini 1.5 Flash**.

This ensures that the advice isn't just "Early Blight"—it's "Early Blight detected in Pune during high humidity; skip watering and use Copper Oxychloride."

hybrid_pipeline.py
async def analyze_crop_health(image, context):
# 1. Fast Edge Inference (10ms)
visual_pred = mobilenet.predict(image)

# 2. Safety Guardrails
if visual_pred.confidence < 0.6:
return Security.flag_uncertainty()

# 3. GenAI Contextualization (1.2s)
advice = await gemini.generate(
role="agronomist",
prompt=f"Diagnosed {visual_pred.disease} in {context.weather}..."
)

return Response(diagnosis=visual_pred, advice=advice)

Beyond Diagnosis

Spraying Index

Real-time wind & rain tracking to advise the perfect time to spray.

Crop Calendar

Dynamic timeline tracking growth stages from sowing to harvest.

Scan History

Local-first storage of all diagnoses for long-term farm tracking.

Back to HomePart of the 2026 Portfolio