AGRINEXO GAIA

Gemini AI assisted Agroecology.

Comment

Inspiration

The inspiration for AGRINEXO GAIA stems from a critical paradox in modern global development: while we live in an era of unprecedented data availability—from hyperspectral satellite imagery to hyper-local climate sensors—the majority of the small and medium-scale farmers, land stewards, and technicians are still making decisions based on intuition or outdated historical precedents. In the face of a changing climate, where "normal" weather patterns no longer exist, intuition may not be enough.

We observed that high-end precision agriculture tools are often locked behind unaffordable subscription fees, complex GIS (Geographic Information System) interfaces, and "black box" algorithms that offer data but no wisdom. The goal was to build a bridge between scientific data and climate models and the operational needs of small and medium-scale farmers.

We were inspired by the concept of an agroecosystem as a living ecosystem rather than a factory. This requires understanding not just a single crop, but the landscape context: the prevailing trees, the local wildlife, the soil, and the specific micro-regional climate.

What it does

AGRINEXO GAIA is an AI-native agroecological intelligence platform. It transforms complex geospatial and scientific data into actionable, conversational insights.

  • Intelligent Geospatial Analysis: Users can draw custom "Field Areas" on an interactive Leaflet map. The system doesn't just calculate area; it triggers a multi-faceted environmental "audit" of that specific polygon.
  • Satellite Health Monitoring (NDVI): The uses NDVI (Normalized Difference Vegetation Index) data. It visualizes plant vigor through dynamic GeoJSON overlays, allowing users to see exactly where a crop is thriving or under stress.
  • The "Six-Decade" Climate Context: One of its most powerful features is the comparison of "Climate Normals." It compares current climate data (the last 30 years) against historical referentials (the 30 years prior, dating back to 1966). This highlights anomalies in temperature and precipitation.
  • Soil Water Balance Modeling: Using FAO-56 standards, the app models Soil Available Water (SAW), Total Available Water (TAW), and Readily Available Water (WRA). It tells the user not just if it rained, but how much water the soil actually retained for the plants.
  • Landscape Context Identification: Using the Gemini "Google Search" grounding tool, users can click any point on the map to receive a comprehensive report on the local ecosystem. This includes prevailing tree species, Köppen-Geiger climate classifications, wildlife prevalence, and even conservation statuses.
  • Computer Vision Weed Analysis: Users can upload photos taken in the field. Using Gemini's multimodal capabilities, GAIA identifies weed species, assesses their prevalence (low/medium/high), and explains their specific impact on the target crop.
  • Conversational Logic (Agrinexo G): The AI assistant isn't just a chatbot; it is a "controller." It has "Function Calling" capabilities that allow it to autonomously decide when to open the analysis modal, which field to look at, and how to interpret the charts for the user.

How we built it

The technical stack was chosen to prioritize speed, scientific accuracy, and a "premium" user experience.

  • Frontend Architecture: Built with React 19 and TypeScript. We opted for a strict typing system to handle the complex, nested data structures returned by agricultural sensors and climate models.
  • The Brain (Gemini API): We utilized the Gemini 3 Flash Preview model. We leveraged Function Calling to allow the model to interface with our custom agronomical database. We also implemented System Instructions to give "Agrinexo G" its specific persona as a botanical expert.
  • Mapping & GIS: Leaflet serves as our map engine. We used react-leaflet to manage the lifecycle of polygons and markers. Complex GeoJSON processing allows us to render NDVI heatmaps directly on the client side with high performance.
  • Data Visualization: We used Chart.js with a modular approach. We built custom plugins, such as a "Vertical Line" plugin to mark "Today" on historical/forecast timelines, helping users distinguish between recorded data and predicted models.
  • Information Grounding: We integrated the Google Search tool within the Gemini configuration. This ensures that when a user asks about a specific micro-region in Portugal or a valley in the Andes, the AI isn't hallucinating; it is reading current, localized landscape data.
  • Backend & API: The application interfaces with the eco.agrinexo.com API, a specialized endpoint that provides processed satellite indices, irrigation models, and historical weather data.

Built With

Share this project:

Updates