// Vadodara, India · open to interesting problems
AI Engineer specializing in computer vision, generative AI and agentic systems, taking deep learning models from research to production, from edge-deployed detection pipelines to LLM-powered multi-agent workflows. Research intern at IISc Bangalore & IIT Kharagpur · CODS Agentic AI Challenge winner · co-author of DynAMO (arXiv) · B.Tech AI & DS '27.
Systems that survived contact with the real world.
End-to-end CV system watching live station feeds for platform intrusions, yellow-line violations and track obstructions. I collected the video data on real platforms, annotated it, fine-tuned YOLOv8x, and squeezed it onto the edge with TensorRT + DeepStream, with ROI-based automated alerts.
Modeling entire railway systems as graphs: centrality, community detection, and graph learning to predict node importance and connectivity behavior.
Vehicle detection, tracking and behavior analysis over large-scale video for real-time incident detection and smart mobility.
Fault-tolerant agentic system for realistic industrial task execution that won the CODS Agentic AI Challenge, presented at the 13th CODS-COMAD International Conference.
Retrieval-augmented generation over PDFs and the web: retriever + generator pipeline with LangChain and FAISS, wrapped in a clean Streamlit UI.
Adaptive signal control from live vehicle density, emergency-vehicle green corridors, and a real-time Firebase-backed dashboard.
Agent-based decision support for farmers: hyperlocal crop advice and weather, designed for simplicity in rural use.
Real-time rider detection, helmet check and number-plate OCR. When a helmetless rider is spotted, the system crops the evidence, reads the plate, and auto-emails a violation log with time and location.
Every stop on the line so far.
Computer vision for intelligent traffic surveillance: detection, tracking, and behavior analysis on large-scale video.
Railway network analysis with graph neural networks and complex network science.
Developing and evaluating large language models: reasoning workflows and LLM evaluation.
Built the real-time station anomaly-detection system from scratch: data collection on platforms → annotation → YOLOv8x fine-tuning → edge deployment.
Statistical analysis of track-defect data across 21 metro sections: EDA, Bayesian networks, clustering. Honored with a stipend.
India's railway university, Vadodara. Where the trains and the tensors met.
Tools I reach for, ordered by muscle memory.
Receipts.
Robust agentic system on AssetOpsBench; invited talk at the 13th CODS-COMAD International Conference.
National-level AI for mobility at IISc Bangalore.
Decentralized AI for transparent data management.
Honored for the Blue Line track-defect analysis work.
Organized and coordinated technical sessions, 2025.
When I'm not training models, I'm out with a camera chasing birds, light, and new places. Photography and travel keep my eyes calibrated.
// next stop
Research collabs, internships, or a hard CV/agents problem. My inbox is open.