Projects
Research-driven projects bridging agentic AI, climate science, and remote sensing.
Agentic Climate Risk Intelligence System
Multi-agent RAG pipeline ingesting 100k+ climate datasets (ERA5, CMIP6, Sentinel-2 LULC) into a vector store, enabling natural language querying of physical risk scores and emissions trajectories. Bayesian inference layer for uncertainty quantification across asset classes, LLM-based agents for autonomous ESG data validation, ReAct-style agentic loops for TCFD/EU Taxonomy compliance, Monte Carlo stress testing for derivatives positioning.
Satellite-Based Exploration Platform
In-house geospatial application built for a geo-exploration division, replacing drone/on-site surveys with open-source satellite data to track vegetation regrowth and oil slicks. Vision Transformer models achieving 92%+ accuracy on multi-spectral imagery across 400+ reclamation sites.
Photogrammetric Species Identification
Photogrammetric species identification pipeline using Sentinel-1 C-band SAR and Sentinel-2 optical data to classify 17+ tree species in Indian forests. GDAL with Python for band-specific data fusion, Random Forest and SVM for refined multi-spectral classification.
Deforestation Detection — M.Tech Thesis
Explored the potential of Sentinel-1 SAR and Sentinel-2 optical satellites for large-scale deforestation detection at IIT Indore's Earth Observation Lab. Multi-temporal analysis combining radar backscatter and spectral indices for change detection in tropical forest cover.
Automated GIS Pipeline — NYU Stern / Project Mumbai
Automated GIS pipeline using Python, GDAL, and PostGIS for NYU Stern's Project Mumbai, cutting 1,200+ manual hours/year. GANs to enhance legacy satellite imagery from 1997+ for remote road detection. Cloud-native flood risk analysis with Monte Carlo simulations on GCP.
Urban Analytics & COVID-19 Economic Impact
Quantified COVID-19's economic impact via nightlight drop analysis using NASA VIIRS data (–37.2% in Delhi). SAR-based urban expansion models across 1,000+ Sub-Saharan towns using PyQGIS/C++, with optimized PostGIS queries for high-throughput spatial analysis.