Research Engineer

MetaFusion · Noida, India

• Built end-to-end fine-tuning and evaluation pipelines for VLM and LLM-based surveillance captioning and attribute prediction in PyTorch. Designed attribute-specific cross-attention layers, profiled and optimised bottlenecks, and fine-tuned Microsoft Florence-2, improving F1 score from 84% to 91%.
• Shipped an intelligent traffic management system integrating multi-violation detection, vehicle recognition, license plate OCR, and CLIP-based attribute prediction, deployed as containerised microservices on GPU-backed infrastructure.
• Owned production rollout across Gandhinagar, Vizag, and Maharashtra with Gradio interfaces for real-time monitoring; built automated benchmarking for quality, latency, and throughput and achieved over 50% end-to-end throughput improvement.

Conference Reviewer

CVPR, ECCV, NeurIPS

Served as a reviewer for AI conferences including CVPR, ECCV, and NeurIPS.

Student Ambassador

AE Global Summit on Open Problems for AI

Selected as a Student Ambassador for the AE Global Summit on open problems in AI. The summit brings together researchers from DeepMind, NVIDIA, and other leading institutions to tackle foundational challenges in AI.

Teaching Assistant

Principles of Computer Systems - IIT Jodhpur · Instructor: Dr. Angshuman Paul

Assisted in teaching a core systems course covering CPU design, memory hierarchy, and MIPS. Conducted labs, prepared assignments, and graded exams for 75+ students.

Teaching Assistant

Introduction to Computer Science - IIT Jodhpur · Instructor: Dr. Mayank Vatsa

Supported a foundational course covering C++, Python, and data structures. Designed tutorials and evaluated coursework for 200+ first-year students.