From advice to action on WhatsApp.

AgriNexus AI gives smallholder farmers fast, local-language agronomic guidance — then follows up until the farmer confirms action. Built on AWS Serverless + Amazon Bedrock (RAG + Vision) with production observability.

Pilot enquiries (email) Try on WhatsApp View code

Tip: Use the email button for pilots/partners (NGO, input company, or government program).

Lambdas deployed (prod)

144

EARS requirements traced to code

80%

Test-to-code ratio (repo metric)

~$0.54

Per farmer / year @ 10K (modeled)

Demo (3 minutes)

What it does

Text / voice / photo queries via WhatsApp (no app install).
Grounded answers via Bedrock Knowledge Base RAG (FAO/ICAR sources).
Vision diagnosis with deterministic gates + structured schema.
Closed-loop nudges with T+24h / T+48h follow-ups until DONE/EXPIRED.

Why this matters

Farmers don’t lose crops because advice doesn’t exist — they lose crops because advice arrives after the window closes.
Extension capacity is limited, but WhatsApp is already in farmers’ hands.
The differentiator isn’t “answer quality” alone — it’s follow-through.

How it works (high level)

WhatsApp → webhook → SQS FIFO → processor
Deterministic gates (rate limit, heuristics, relevance/quality)
Bedrock RAG (text) + Vision (images) + Polly/Transcribe (voice)
Weather-triggered nudges → reminders → DONE/NOT YET detection

Architecture diagrams Metrics & monitoring