AI Researcher and Computer Vision Engineer with 6+ years of experience in building scalable vision systems across industry and academia. Currently focused on learning structured world representations through semantic scene graphs and multimodal encoders for downstream tasks such as autonomous systems and healthcare applications.
Working on multi-robot perception and representation learning using semantic and geometric scene understanding. Designing graph-based world models from multi-view camera inputs for robot localization, calibration, and interaction.
Developed segmentation algorithms for large satellite and volumetric medical datasets on Jetson edge platforms. Presented at ICASSP 2025.
Improved player tracking accuracy from 85% to 98% using temporal smoothing. Designed robust Jersey OCR (TensorFlow) with 90%+ accuracy.
Built CV APIs handling 2.5M+ req/day. Designed multi-camera vehicle tracking for 300+ CCTV cams. Built lighting-invariant color detection.
Designed a sensorless force estimation model utilizing joint currents for the uFactory Lite6. Engineered autonomous tabletop object sorting and AprilTag camera calibration pipelines.
Built ResearchBuddy, a RAG-Based QA pipeline using Gemini-2.5 Flash and LangChain. Developed OPD Assistant Medical Record System with PGIMER to convert unstructured raw data to meaningful records.
Focused on learning structured world representations through semantic scene graphs and multimodal encoders.
Improved player tracking accuracy from 85% to 98% via temporal smoothing. Designed robust Jersey OCR (TensorFlow).
Built CV APIs handling 2.5M+ req/day. Designed multi-camera vehicle tracking for 300+ CCTV cams.
Designing graph-based world models from multi-view camera inputs for robot localization, calibration, and interaction.
Developed segmentation algorithms for large satellite and volumetric medical datasets on Jetson edge platforms.
Sensorless Force Estimation (Lite6), Autonomous Tabletop Object Sorting, AprilTag Calibration Pipeline, ROS2 & Gazebo Sims.
ResearchBuddy: RAG-Based QA using Gemini-2.5 Flash. OPD Assistant Medical Record System built with PGIMER.
ACM HRI 2026, Edinburgh. Transforms HRI from descriptive analysis to a priori verification.
ICASSP 2025. Optimized computer vision for edge systems under resource constraints.
NeuroHRI 2025. CVPR MedSegFM 2025 (Ranked 4th/210). Indian Patent App (202411050462) for DL Training Methods.
Department of CSE (Human Machine Interaction Lab).
Thesis: Localized Perception for Constrained Vision Systems.