Geoff (GEOspatial Fact Finder) is a geospatial AI that turns natural language into spatial queries and displays results on a web map

Winner of FIAP'S Global Solution 2025.1 Challenge. This repository contains the architecture for a multi-agent system where five autonomous "Guardians" work in synergy to predict, manage, and respond to complex events like wildfires, epidemics, and infrastructure failures.

TN-Epic is the world’s first State-Scale Experience OS, transforming Tamil Nadu into a playable AR Meta-Game. By turning monuments into levels and merchants into NPCs, we solve the 80% revenue leakage and ₹1,200Cr sanitation crisis. Built with ARCore, YOLOv8, and Java Spring Boot, it gamifies civic duty for 286M annual tourists.

साहित्याच्या पुणेरी पाट्या - Where Marathi literature meets AI. Explore 47+ locations and 60+ quotes across Pune through an interactive 3D map powered by vector search and iconic 'Puneri Patya' design.

Ready-to-use MCP extensions for Gemini CLI and Claude Desktop - neighborhood intelligence, school ratings, flood risk, air quality, and more.

Geospatial deep learning pipeline for detecting illegal mining structures and machinery in satellite imagery using YOLO and spatial post-processing.

FRA Atlas AI is an intelligent forest rights digitization and monitoring system that automates document verification via OCR, classifies land using CNN & ISRO Bhuvan data, and powers a LangChain-based Decision Support System for transparent, data-driven governance.

Binary classification experiments to interpret Google AlphaEarth Foundation embeddings across ESA WorldCover land cover classes. Part of the study "What on Earth is AlphaEarth?" — 130,000+ experiments using Random Forest, XGBoost, LightGBM and progressive ablation across 64 embedding dimensions.

OpenMMLab-based remote sensing segmentation pipeline for binary water extraction using SegFormer, ensemble inference, and test-time augmentation.

Geospatial AI for land use/land cover change detection correlated with population dynamics using WorldPop COGs, satellite imagery segmentation, and Next.js visualization

Geospatial AI system for detecting and classifying surface changes using Sentinel-2 imagery, PyTorch Siamese networks, and Google Earth Engine.

AI-Powered Climate Vulnerability Assessment for Transportation Infrastructure using Google's Satellite Embeddings

AI-powered wildfire damage detection using IBM/NASA Prithivi-EO-2.0 foundation models and hybrid spectral analysis for high-precision burn mapping.

A comprehensive, end-to-end geospatial pipeline for wildfire damage assessment. Integrates Google Earth Engine (GEE) data harvesting with NASA-IBM Prithvi EO 2.0 (Vision Transformer) fine-tuning to achieve a +50.58% Macro F1 improvement in burn scar severity classification.

Nairobi-first edge AI simulation for urban flood resilience. Orchestrating multimodal agents (Gemini 2.5/Gemma 4) to navigate the Mbagathi Basin during infrastructure failure. Offline-first, spatial-aware, and mission-critical.

PineGuard: A geospatial AI platform for Pine Wilt Disease (PWD) detection. Integrates 40-year Sentinel-2 spectral analysis, land-cover masking, and regression modeling to identify infection epicenters and project 20-year strategic risk.