Production-Grade AI Platform (Standalone Release)

Sanjivani 2.0

The Next-Gen AI Agronomist Platform. Hybrid CNN + Gemini 1.5 Flash architecture that understands your crops, speaks regional dialects, and optimizes spraying schedules.

MobileNetV2 Edge
Gemini 1.5 Flash

Hybrid Intelligence

Dual-Mode AI

Deterministic vision for 95% accuracyin <10ms, coupled with LLM contextual reasoning to synthesize weather, soil, and regional spray advice.

< 10ms
Edge Prescreen Latency
3 Langs
English / Hindi / Marathi
Next.js 14
FastAPI Microservices
Gemini 1.5 Flash
TensorFlow Lite
Cloudflare Pages
Render
Firebase
Product Evolution

The 2.0 Architectural Shift

Evolving from a single offline diagnostic tool (1.0) into a full microservices farm management ecosystem (2.0).

Sanjivani 1.0 (Edge Vision Tool)

  • Single Monolith (Flask API)
  • Manual Server Deployment
  • English Text Only
  • Fixed Image Disease Classification

Sanjivani 2.0 (Platform Suite)

  • Decoupled Microservices (Next.js 14 + FastAPI)
  • Automated CI/CD on Cloudflare Pages & Render
  • Native Marathi, Hindi & English Voice UI
  • Contextual Spray Index & Farm Growth Suite
Hybrid AI Engine

Hybrid CNN + LLM Orchestration

Combining ultra-fast deterministic edge vision with LLM contextual reasoning.

01 // STAGE

10ms Edge Prescreen

MobileNetV2 handles initial visual scan. If confidence is below 60%, safety guardrails reject non-plant or noisy images.

02 // STAGE

Real-time Context Fetch

FastAPI microservice fetches hyper-local humidity, wind speed, and precipitation data for the farmer's GPS location.

03 // STAGE

Gemini 1.5 Flash Synthesis

LLM contextualizes the visual diagnosis with weather risks (e.g. high humidity warning for Copper Oxychloride application).

04 // STAGE

Regional Voice Output

Generates spoken Marathi/Hindi advice so farmers without literacy barriers receive immediate step-by-step guidance.

Engineered for
Resilience.

Sanjivani 2.0 moved beyond static classification. Instead of returning raw labels ("Early Blight"), the Hybrid AI Pipeline evaluates environmental risks before generating remedies.

If high humidity and rain are forecast, the system advises delaying spray applications to prevent chemical runoff—saving farmers money and protecting soil health.

Gemini 1.5 Flash LLM

Contextual agronomist prompt engineering with weather payload injection.

Multilingual Translation Engine

Native Marathi & Hindi localization with Web Speech API audio playback.

hybrid_pipeline.py
async def analyze_crop_health(image, context):
  # 1. Fast Edge Prescreen (<10ms)
  visual_pred = mobilenet.predict(image)
  if visual_pred.confidence < 0.6:
    return Security.flag_uncertainty()

  # 2. Fetch Hyper-local Weather Payload
  weather = await weather_service.get(context.lat, context.lon)

  # 3. Gemini 1.5 Flash Agronomist Context
  advice = await gemini.generate(
    role="agronomist",
    lang=context.language,
    prompt=f"Diagnosed {visual_pred.disease} in {weather.humidity}% humidity..."
  )
  return Response(diagnosis=visual_pred, advice=advice)

Beyond Diagnosis

Spraying Index

Real-time wind speed & rain tracking to advise the safest hours for spray application.

Crop Growth Calendar

Dynamic timeline tracking growth milestones from sowing to harvest with preventative tips.

Scan History & Trends

Cloud database sync across mobile devices to track farm recovery over multiple seasons.