Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/sdarionicolas-boop/AgroIA-RAG/llms.txt

Use this file to discover all available pages before exploring further.

AgroIA is an automated agronomic diagnostic system built for Latin American agriculture. It takes a GPS point or shapefile as input and — in under 60 seconds — delivers a precise field polygon, a multi-variable AgroIA Score (0–100), historical NDVI analysis, and a PDF report with interactive map. All processing, including the LLM, runs locally to preserve full data sovereignty.

Quick Start

Get AgroIA running with Docker in minutes and run your first crop analysis.

Architecture

Understand the end-to-end pipeline from GPS input to RAG-powered reports.

AgroIA Score

Learn how the 0–100 score is computed from Vigor, Stability, Cleanliness, and Climate.

API Reference

Explore REST endpoints for ingestion, querying lots, and bulk GeoJSON upload.

How it works

AgroIA automates the full agronomic diagnostic lifecycle — from field boundary detection to expert natural-language consultation.
1

Delineate your field

Provide a GPS point or shapefile. SAM (Segment Anything Model) analyzes Sentinel-2 imagery to automatically generate a precise field polygon. No manual digitizing required.
2

Run the analysis pipeline

The local pipeline pulls 6 years of Sentinel-2 NDVI data from Google Earth Engine and climate data from NASA POWER, then calculates the AgroIA Score with anomaly filtering via IsolationForest.
3

Get your reports

Receive a PDF agronomic report and an interactive HTML map with A/B/C zone classification. Results are automatically ingested into the vector database.
4

Consult the RAG agent

Ask natural-language questions about any field — “What is the stability trend for lot X?” — and the Gemma 3 model retrieves relevant historical context from pgvector to answer.

Key capabilities

SAM field delineation

75% precision validated against INTA Balcarce reference data. GPS point to GeoJSON polygon in seconds.

6-year NDVI history

Sentinel-2 SR time series via Google Earth Engine with automatic cloud/error filtering.

Batch processing

Process thousands of fields from a single GeoJSON file with one command.

Local RAG engine

pgvector + Ollama (Gemma 3) runs entirely on-premise. No data leaves your infrastructure.

Streamlit dashboard

Visual explorer for all lots with score rankings, NDVI charts, and map overlays.

Telegram bot

Query agronomic data and run diagnostics directly from a Telegram chat.

Validation metrics

AgroIA has been validated against real-world datasets from TAYPE Siniestros (313 points) and INTA Balcarce (454 points):
MetricValue
Hit Rate (TAYPE)85.6%
Hit Rate (INTA Balcarce)74.9%
Average SAM Score0.962
Average area error vs. manual9.8%
End-to-end processing time~60 seconds
AgroIA requires a Google Earth Engine project ID and active credentials for the full satellite pipeline. See Environment Configuration for setup details.

Build docs developers (and LLMs) love