Edge-AI on NPUs
Quantization, packaging, and optimized inference on automotive and industrial NPUs — TDA4VM, Hailo-15, R-Car, Rockchip, Mobilint Aries, DeepX M1, and more.
Open to engineering roles
Usman Asim
Computer vision, MLOps & edge-AI engineer building deep-learning systems where hardware, latency, and reliability actually matter — from driver monitoring to medical imaging.
7+
Years in CV / edge AI
7
Engineers led
10×
Faster inference
6+
NPU platforms
01
What I do
I build, optimize, and ship deep-learning vision systems on embedded NPUs and in the cloud — across driver & occupant monitoring, road perception, industrial safety, and medical imaging. On TI TDA4VM, Hailo-15, Renesas R-Car, Rockchip, Mobilint Aries, and DeepX M1 targets, I've led a seven-engineer team, built Docker-based MLOps pipelines, and delivered up to 10× faster inference than Python laptop baselines. Published researcher with six peer-reviewed papers.
Driver & occupant monitoring
Face detection, landmarks, gaze, and drowsiness on embedded IR camera systems, aligned to Euro NCAP-oriented DMS requirements.
Reproducible delivery
Docker-based CI/CD for model release consistency, Yocto SDK customization, and reproducible embedded deployment pipelines.
Medical imaging
Segmentation and classification research, including 3D-CNN diagnosis on brain MRI and active-contour methods. Six peer-reviewed papers.
02
Experience
AI Consultant part-time
Evismart · AI Product Consulting
Advised an in-house AI product team on a dental-scan segmentation product, end to end.
- Guided computer-vision model design, optimization, and deployment strategy for dental-scan segmentation.
- Translated clinical imaging requirements into practical model-development and production-readiness guidance.
AI Research Engineer
ConTiLab · Industrial Safety AI
Brought the iSafeGuard industrial-safety platform onto embedded AI hardware.
- Deployed iSafeGuard YOLOv8 worker-safety models on Mobilint Aries and DeepX M1 NPUs and benchmarked performance across embedded targets.
- Built a remote AI model-management and monitoring system for camera-based safety devices running on embedded NPUs.
Team Lead · Computer Vision & Edge-AI Engineer
DELTAX · Automotive AI
Led seven engineers across model optimization, deployment, and integration for automotive AI programs delivered to Hyundai, Kia, and LG Display.
- Delivered multi-model inference pipelines up to 10× faster than Python laptop baselines across TI TDA4VM, Hailo-15, Renesas R-Car, and Rockchip.
- Customized Yocto-based Linux SDKs and built OpenVX / embedded C/C++ apps for fisheye IR driver & occupant monitoring cameras, integrating in-vehicle sensor-fusion inputs.
- Automated Docker-based CI/CD for releases; shipped models spanning driver & occupant monitoring, surround-view road perception, road-crack detection, in-cabin gesture UI, and factory surveillance.
DEPLOYMENT TARGETS
TI TDA4VM
Hailo-15
Renesas R-Car
Rockchip
Mobilint Aries
DeepX M1
AI Research Engineer
UNOMIC Inc. · Medical AI
Took a medical-imaging classifier from research to a regulated, cloud-native deployment.
- Developed and deployed a cloud-native, containerized 3D-CNN for Alzheimer's-disease classification from 3D brain MRI.
- Supported a Korea MFDS medical-device registration submission under Personal Information Protection Act (PIPA) requirements.
Researcher
AI Vision Lab · Chung-Ang University · Research
First-author IEEE Access work on active-contour and dilated-convolution methods for inhomogeneous image segmentation. Contributed to an NRF-funded brain-stroke image-generation and segmentation project using GANs, and to a National Information Society Agency maritime object-detection dataset.
Research Assistant intern
ARFA Software Technology Park · Research
Built U-Net-based 2D brain-MRI segmentation experiments and ran transfer-learning comparisons across CNN baselines; collected and labeled an offline-signature recognition dataset for downstream verification.
03
Selected projects
Cabin & driver monitoring systems
Driver and occupant monitoring for automotive programs — face detection, facial landmarks, gaze tracking, and drowsiness detection on fisheye IR cameras, deployed as embedded C/C++ pipelines on automotive NPUs under real-time and Euro NCAP-oriented constraints. Client work; visuals are representative.
TI TDA4VM
Renesas R-Car
OpenVX
Embedded C/C++
Cloud-native Alzheimer's detection
Containerized 3D-CNN application classifying Alzheimer's disease from 3D brain MRI — DICOM conversion, hippocampus segmentation, and cloud-side inference — supporting a Korea MFDS medical-device submission.
Active contour model with dilated convolution
First-author IEEE Access work combining active contour modeling with a local dilated convolution filter for robust segmentation of inhomogeneous synthetic, natural, and medical images.
DOI ↗
Docker guide for AI development & deployment
Practical open-source guide to packaging computer-vision workloads with Docker — container concepts, OpenCV inference examples served through Flask, and reproducible build-and-run workflows.
GitHub ↗
Medical-imaging classification toolkit
Final-year MATLAB toolkit for medical-imaging research: dataset loading, preprocessing, handcrafted (HOG, LBP) and deep (AlexNet, ResNet) feature extraction, and comparative classifier evaluation.
04
Education
M.S. Computer Science & Engineering
Chung-Ang University
Thesis: Active Contour Model for Image Segmentation with Dilated Convolution Filter. Research in image segmentation and medical imaging, including 3D-CNN diagnosis on brain MRI. Teaching assistant for Advanced Image Processing. Six peer-reviewed papers across IEEE Access, Springer, and international conferences.
B.S. Computer Science
COMSATS University Islamabad
Final-year project: an automated medical-imaging classification toolkit in MATLAB (dataset loading, preprocessing, handcrafted & deep feature extraction, classifier comparison). Teaching assistant for Object-Oriented Programming, Digital Image Processing, and Neural Networks.
05
Toolkit
Languages
Python
C / C++
C#
Julia
MATLAB
Vision & ML
PyTorch
TensorFlow
OpenCV
ONNX
Caffe2
3D CNNs
Edge & Embedded
Embedded Linux
Yocto
OpenVX
NPU SDKs
Model Optimization
Sensor Fusion
NPUs & Hardware
TI TDA4VM
Hailo-15
Renesas V3H / V4H
Rockchip RK3588
Mobilint Aries
DeepX M1
NVIDIA Jetson / DGX
MLOps & Tooling
Docker
Kubernetes
Kubeflow
CI/CD
Git
Jira / Confluence
Cloud & Data
GCP
Azure
MySQL / SQL
FreeSurfer / FSL / SPM12
06
Publications
2023
Springer · PAA
★ Selected
ViT-PGC: Vision Transformer for Pedestrian Gender Classification on Small-Size Dataset
2022
JMIS
3D-CNN Method over Shifted Patch Tokenization for MRI-Based Diagnosis of Alzheimer's Disease Using Segmented Hippocampus
IEEE · ARACE
Image Generation Network Model Based on Principal Component Analysis
2021
IEEE Access
★ Selected
Active Contour Model for Image Segmentation with Dilated Convolution Filter
DOI ↗
IEEE · ISPACS
Saliency-Based Active Contour Model for Image Segmentation and Region Detection
2020
IEEE Access
Saliency-Driven Active Contour Model for Image Segmentation
DOI ↗
07
Certifications, awards & languages
Certifications
TensorFlow: Advanced Techniques
ML Engineering for Production (MLOps)
Convolutional Neural Networks
IELTS — Band 7
Awards & honors
CAU Young Scientist Scholarship (CAYSS)
Winner — Speed Programming
Languages
English
Urdu
Punjabi
Korean
08
Get in touch
Open to engineering roles, collaborations, and technical conversations in computer vision & edge AI.