At SRC Business Solutions, I build computer vision and machine learning systems for an agroforestry program spanning multiple Indian states. I fine-tuned a vision transformer for 100-species tree classification, reaching 95.2% top-1 accuracy, and deployed it as an ONNX inference service.
I’ve also developed satellite tree detection and canopy mapping across 93 farms, a geospatial soil-profiling pipeline, and tools for detecting duplicate or fraudulent field photos. For tree-height estimation, I combined instance segmentation, depth estimation, and camera geometry.
At BharatGen Consortium, I worked on handwritten text recognition and OCR robustness. My research includes a two-pass transformer that reduced character error rate by 56.1% over baseline, and I co-developed EroPT, a 2,437-image benchmark; my work has appeared in four peer-reviewed papers.
