AgriBridge

AgriBridge

Inspiration

Smallholder farmers face devastating crop losses from diseases, pests, and shifting climate patterns, yet they rarely have timely access to agricultural extension officers. On the other hand, thousands of peer-reviewed agricultural research papers are published every year, but this scientific knowledge remains trapped in academic journals. We created AgriBridge to bridge this gap—instantly turning real-world farmer field queries into structured research intelligence while providing farmers with evidence-backed AI diagnosis.

What it does

AgriBridge is an AI-powered agricultural intelligence platform built on three main modules: Farmer Portal (/farmer): Farmers can type, speak, or upload photo symptoms of infected crops in English or Bengali. Google Gemini 2.5 Flash provides instant visual diagnosis, confidence ratings, and safe remedies. Research Intelligence Hub (/research): Automatically converts farmer queries into structured problem clusters across 14 fixed agricultural categories and links them directly to peer-reviewed research papers fetched live from the OpenAlex API. Features interactive stat cards with search modals. Farm Intelligence (/farm-intelligence): Allows farmers to drop a pin on an interactive Leaflet map, fetching live temperature, rainfall, and solar data from Open-Meteo API to generate customized planting and irrigation advice.

How we built it

Frontend & App Framework: Next.js 16 (App Router), React, TailwindCSS, Lucide Icons, and Leaflet.js maps. AI Engine: Google Gemini 2.5 Flash API (@google/genai) using structured JSON mode for problem metadata extraction and intent classification. Integrations: Open-Meteo Weather API for localized micro-climate forecasts, OpenAlex Works API for scientific literature. Database: Hybrid architecture featuring an active in-memory singleton store with full Supabase PostgreSQL database persistence support.

Challenges we ran into

Greeting vs. Problem Classification: Preventing casual messages (like "Hello" or "Hi") from creating fake disease problem clusters in the research database required fine-tuning Gemini's structured JSON output schema. Dosage Safety Constraints: Designing AI system prompts that deliver helpful agronomic advice while strictly preventing unverified chemical pesticide dosage recommendations. Model Aliasing: Handling evolving Gemini API model identifiers cleanly with automatic fallback loops to guarantee 100% uptime.

Accomplishments that we're proud of

Creating a seamless end-to-end loop where a farmer's voice/image query instantly generates structured research intelligence in the same session. Full bilingual support (English & Bengali) tailored to smallholder farming communities. Building a resilient fallback architecture that handles network failures or missing API keys gracefully with clear developer UI feedback.

What we learned

How to combine multi-modal LLM reasoning with real-time external APIs (Open-Meteo & OpenAlex) for evidence-backed advisories. Effective JSON mode prompt engineering for simultaneous conversational and structured extraction outputs. Designing accessible, intuitive web interfaces for rural users with voice input and clear visual indicators.

What's next for AgriBridge

Offline SMS / Voice Gateways: Enabling IVR and SMS query submission for farmers in low-connectivity rural zones. Satellite Vegetation Indices (NDVI): Integrating Copernicus Sentinel-2 STAC imagery for automated regional crop stress heatmaps. Agronomist Dispatch Integration: Connecting high-priority problem clusters directly with local Sub-Assistant Agriculture Officers (SAAO) for on-ground field visits.