Step by Step
Six steps. One goal. Certainty in 30 seconds.
A complete walkthrough of what happens from the moment a farmer sends a photo to when they receive actionable advice.
Farmer sends a photo
The farmer takes a photo of their crop — damaged leaves, unusual growth, discoloration — and sends it via WhatsApp, SMS, or the app. No training required. No special equipment. Any phone camera works.
Technical detail
Photo ingested via Twilio WhatsApp API or direct app upload. Stored securely in our media pipeline with automatic compression.
Context engine assembles
In parallel with photo processing, our context engine assembles everything we know about this farm's current situation — in under one second.
Technical detail
PostGIS geospatial lookup → GPS-matched weather from Open-Meteo API → soil type from regional database → current crop stage from farm profile → last 30 interactions from history store.
AgroAI analyses with full context
Claude Vision processes the crop photo alongside the full assembled context. Not just 'what disease is this?' but 'what disease is this, given this farm's GPS, current weather, soil type, and crop stage?'
Technical detail
Claude Vision API with structured agricultural reasoning prompt. Context window includes: image + weather JSON + soil profile + crop stage + interaction history summary.
Structured insight generated
The AI generates a structured response with diagnosis, specific action, confidence score, urgency level, and reasoning — formatted for the delivery channel.
Technical detail
Output schema: { diagnosis, action, confidence: 0–1, urgency: 'high|medium|low', reasoning, follow_up, product_name, dosage, timing }
Delivered to the farmer
The response is formatted for the farmer's channel — conversational and clear via WhatsApp, structured via the app — and delivered in under 5 seconds from when they sent the photo.
Technical detail
Twilio outbound WhatsApp message. App push notification + in-app display. SMS fallback with condensed response.
Interaction logged for future context
The full interaction — photo, context, diagnosis, action taken — is logged. Every future interaction from this farm benefits from richer history. The system gets smarter with every use.
Technical detail
PostgreSQL write: interaction record linked to farm_id. History index updated. Anonymised record flagged for AI training pipeline if farmer has not opted out.
Every Response
Structured. Actionable. Explainable.
Every PreciAgro response contains the same structured output. Not just a diagnosis — a complete action plan with confidence scoring and reasoning.
- What we detected (diagnosis)
- What to do, when, how much (action)
- How confident we are (0–100%)
- How urgent the action is
- Why we recommend this (reasoning)
- What to watch for next (follow-up)
{
"diagnosis": "Fall Armyworm + Grey Leaf Spot",
"action": "Apply Emamectin Benzoate 5% SG
10g/15L + Mancozeb 50g/15L",
"timing": "Today before 4pm",
"confidence": 0.94,
"urgency": "high",
"reasoning": "Lesion pattern consistent with
FAW feeding + GLS spores active in
current humidity (87%). Weather window
closes at 5pm.",
"follow_up": "Re-inspect in 5 days.
Watch upper canopy expansion.",
"location": "Mashonaland West",
"weather_context": {
"temp": 28,
"humidity": 87,
"rain_forecast": "tomorrow"
}
}Performance
Target benchmarks at scale
<5s
End-to-end response time
<1s
Context assembly
94%
Accuracy target
99.5%
Uptime target