Open to opportunities

Arman Ayub

@armanayub

Message

Computer vision and edge-AI research intern, building privacy-aware pose estimation and deployable ML systems.

India

Message

What I'm looking for

I’m looking for an opportunity where I can build and deploy computer vision/edge-AI systems end-to-end—combining research rigor (compression, quantization, data strategy) with practical deployment on real hardware.

I’m a Computer Vision and edge-AI research intern focused on turning high-performing models into fast, privacy-aware deployments. At Ritsumeikan University, I designed a full edge-deployed thermal human pose estimation pipeline for elderly care monitoring, targeting FLIR Lepton 3.5 performance on NVIDIA Jetson TX2.

I built 9 dataset configurations from 4 thermal sources (141,154 annotations) and ran systematic experiments across SimpleBaseline, HRNet-W32 (from scratch), and ViTPose-B. Using D6 (HRNet-W32), I achieved 92.96% mAP on OTP2, 82.31% mAP on 160×120 low-resolution, and validated latency of 63 ms/frame (~15.3 FPS) with TensorRT.

I also explored model compression and deployment reliability—analyzing INT8 quantization vs. unstructured pruning and finding that 30% pruning + INT8 exceeded direct INT8 by +0.91 mAP at identical model size and latency. From there, I extended the same end-to-end mindset across internships and projects, including MediaPipe-based pose pipelines, KAN deployment on Android via ONNX Runtime, and NLP systems with LangChain and Gemini.

Experience

Work history, roles, and key accomplishments

Current

Edge Pose Estimation Intern

Current

Ritsumeikan University

Mar 2026 - Present (3 months)

Designed an edge-deployed thermal human pose estimation pipeline for privacy-aware elderly care monitoring on NVIDIA Jetson TX2, achieving up to 92.96% mAP on OTP2 and 63 ms/frame using TensorRT. Built 9 dataset configurations (141,154 annotations) and evaluated pruning+INT8, identifying a setup that improved accuracy by 0.91 mAP at the same model size and latency.

NVIDIA Jetson TensorRT HRNet W32 YOLO Detector Integration INT8 Quantization

Computer Vision Intern

CoachBuddy AI

Jun 2025 - Sep 2025 (3 months)

Built a MediaPipe-based badminton shot classification pipeline by extracting pose keypoints from match videos and constructing a labelled dataset. Implemented pose normalization and augmentation, visualized pose trajectories, and developed a cosine-similarity pose scoring system for frame-level shot identification.

Computer Vision MediaPipe Video Dataset Augmentation Cosine Similarity Scoring Model Evaluation Mediapipe

Salesforce Intern

Novigo Solutions Pvt. Ltd.

Jun 2024 - Jul 2024 (1 month)

Designed and deployed Salesforce CRM applications using Trailhead while collaborating with cross-functional teams across the SDLC. Supported work from requirements gathering through deployment.

Salesforce CRM Trailhead Development SDLC Requirements Gathering Deployment

Education

Degrees, certifications, and relevant coursework

NMAM Institute of Technology – Nitte (Deemed to be University)

Bachelor of Technology, Electronics & Communication Engineering

2022 -

Grade: CGPA: 7.87/10.0 (after 7 semesters); SGPA: 9.14 (Sem 6); Sem 1: 7.14; Sem 7: 7.87

Activities and societies: Relevant coursework: Machine Learning, Artificial Intelligence, Data Structures & Algorithms, Digital Signal Processing, Computer Networks & Cyber Security, VLSI Design, Microcontrollers, IoT, OOP in Java.

Pursuing a Bachelor of Technology in Electronics & Communication Engineering with an expected graduation of June 2026. Maintained a CGPA of 7.87/10.0 (after 7 semesters) with an upward trend from 7.14 in Sem 1 to 7.87 in Sem 7.