Engineer building production computer vision, multimodal AI, and ML systems.

I work across model development and deployment, from training and evaluation to edge inference, backend services, and production observability. I hold a B.Tech in Computer Science and Engineering from IIIT Delhi.

Based in Bengaluru, India.

Email | Phone: 9811370971

• Production computer vision: detection, tracking, model distillation, failure analysis, TensorRT optimization, and Jetson deployment
• Multimodal and generative AI: vision-language models, diffusion pipelines, LoRA/PEFT fine-tuning, and video/audio systems
• ML systems: data and evaluation pipelines, APIs, containerized deployment, monitoring, and distributed infrastructure
• Applied research: video anomaly detection, monocular speed estimation, traffic-sign understanding, and tree-search methods for LLM reasoning

• Reduced a Qwen2-VL video-captioning pipeline from roughly 5 hours to 10 minutes, a 96% latency reduction
• Improved an edge detector's recall from 81 to 94 through an end-to-end teacher-student distillation pipeline
• Raised a hand detector's F1 score from 70 to 93 through systematic failure-mode analysis and dataset iteration

Most of my production work is private. The repositories below focus on research, prototypes, and systems that I can share publicly.

A research-backed lecture-video evaluation pipeline that combines nonverbal-cue analysis, long-context VLM review, evidence-linked coaching reports, and a Streamlit interface. The repository includes Docker setup, evaluation artifacts, and a thesis-style technical report.

An anomaly-detection and surveillance system evaluated on UCF-Crime, with a browser-based monitoring interface, real-time alerts, and replayable evidence. The project explores lightweight backbones, efficient attention, quantization, and adaptive frame processing.

A reproduction of Feng et al.'s tree-search framework for LLM decoding and training, using Monte Carlo Tree Search and learned value estimates to guide reasoning.

A React, Express, and MySQL booking system focused on concurrency and transaction safety, including time-limited reservations, row locking, atomic confirmation, rollback handling, and inventory tracking.

• ML and computer vision: PyTorch, OpenCV, Hugging Face, DeepSpeed, YOLO, RF-DETR, TensorRT, MLflow
• Generative and multimodal AI: Qwen-VL, LoRA/PEFT, diffusion models, LangGraph, RAG, Pinecone
• Systems and infrastructure: Docker, Kubernetes, FastAPI, Flask, Redis, Cassandra, Prometheus, AWS, RunPod
• Languages: Python, C++, JavaScript, TypeScript, SQL, Bash

• B.Tech in Computer Science and Engineering, IIIT Delhi, 2022-2026, CGPA 8.14
• Undergraduate research in monocular speed estimation and traffic-sign detection for large-scale roadway video