Computer Vision Systems
Detection, tracking, OCR, ANPR, face recognition, pose analysis, and multi-camera video analytics.
About me
Machine Learning Engineer specializing in production Computer Vision and Edge AI.
I design and deploy reliable computer-vision systems for safety, security, logistics, manufacturing, and retail. My work spans the full delivery path—from data and model development to tracking, temporal decision logic, APIs, evidence generation, and edge inference.
Alongside production engineering, I conduct peer-reviewed research in computational healthcare. That research discipline shapes how I evaluate models, communicate uncertainty, and build systems that remain trustworthy outside a controlled experiment.
- 28
- CV/ML works Industry, R&D, research, and academic work
- 11
- Industry deployments Deployed projects from ANPR through Mobil
- 3
- Research papers One journal article and two preprints
- CV + Edge AI
- Engineering focus Models, systems, and deployment
Core capabilities
Engineering from model to production
Model Engineering
Dataset curation, training, evaluation, error analysis, and reproducible experimentation with PyTorch and TensorFlow.
Edge AI Deployment
Real-time inference on NVIDIA Jetson, TensorRT optimization, GPU Linux systems, and resilient RTSP processing.
MLOps & Data Engineering
Annotation workflows, versioned data, experiment tracking, validation, monitoring, and delivery pipelines.
Systems Engineering
FastAPI services, Docker, event queues, temporal rules, evidence generation, telemetry, and retryable integrations.
Research & Evaluation
Applied research, subject-aware validation, explainability, signal processing, and peer-reviewed publication.
Selected achievements
Selected engineering and research work
Representative contributions across industry implementation and applied research.
Production CV across industries
Built systems for banking, industrial safety, logistics, manufacturing, and retail environments.
FocusMultiple real-world domainsRead case studyISO 6346 container OCR
Engineered an edge pipeline for container-code recognition with validation and telemetry integration.
FocusCheck digit + telemetryRead case studyBangla ANPR system
Developed track-centric plate recognition with oriented detection, OCR validation, voting, and deduplication.
FocusDetection + OCR + trackingRead case studyMulti-camera video analytics
Architected tracking pipelines with temporal rules, evidence queues, and reliable downstream delivery.
FocusReal-time operational workflowsRead case studyJetson & TensorRT deployment
Deployed and optimized computer-vision workloads for NVIDIA Jetson and multi-GPU Linux systems.
FocusEdge and GPU production stacksRead case studyJournal of Voice publication
First-authored peer-reviewed research on Bengali voice-based mental-health assessment.
FocusJournal of Voice · 2024Read case studyHow I build
Reliable delivery, end to end
- 01
Data
Collect, label, audit, and version representative datasets.
- 02
Modeling
Train, validate, and analyze failure modes against strong baselines.
- 03
Systems
Connect inference to tracking, decision logic, APIs, and evidence.
- 04
Deployment
Ship to edge or GPU infrastructure with monitoring and recovery paths.
Research foundation
Evaluation grounded in evidence
My research covers Bengali voice-based mental-health assessment, spectrogram learning, Vision Transformers, explainability, and noise-robust evaluation. It strengthens the way I design experiments and validate production decisions.
Selected training & community
Continuous learning and service
Training & certifications
- Tools for Data ScienceIBM · Coursera · Sep 2024
- Generative AI Productivity SkillsMicrosoft & LinkedIn · Aug 2024
- Prompt Engineering for Generative AILinkedIn · Jul 2024
- Brain-Computer InterfacePantech.AI Academy · Professional training
Community & leadership
- Red Crescent Youth VolunteerBangladesh Red Crescent Society · Feb 2021 – Aug 2022
- Vice President, Arts and CraftsBAF Shaheen College Dhaka Science Club · 2017 – 2018
Experience
Production engineering & applied research
End-to-end ownership across model development, deployment, and evaluation
Machine Learning Engineer
Bondstein Technologies Ltd.Build and deploy production computer vision systems for enterprise safety, security, logistics, retail, and manufacturing workflows. Own the lifecycle from dataset and model development through RTSP processing, tracking and temporal rules, evidence generation, API integration, and NVIDIA Jetson or GPU deployment. Representative work includes ISO 6346 container-code OCR, Bangla ANPR, factory SOP and PPE monitoring, multi-camera banking surveillance, retail analytics, and industrial visual inspection.
PythonYOLO11YOLO-OBBPaddleOCROpenCVBoT-SORTByteTrackInsightFaceFAISSTensorRTNVIDIA JetsonFastAPIDockerSQLite
Research Assistant
4IR Research Cell, Daffodil International UniversityConducted applied research in Bengali voice-based mental-health assessment and interpretable medical AI under Dr. Md. Taimur Ahad. Developed spectrogram-based CNN and Vision Transformer experiments and studied model interpretation. Continued research beyond this appointment explores recurrent hybrids and self-supervised speech representations. First author of a peer-reviewed Journal of Voice article and contributor to continuing work on robust, subject-independent evaluation.
PyTorchTensorFlowVision TransformersDenseNetGRUwav2vec 2.0WavLMWhisperLibrosaGrad-CAMLIMESHAPWeights & Biases
Education
Academic background
Formal training supporting applied machine learning and systems engineering
BSc. in Computer Science and Engineering
Daffodil International University (DIU), Dhaka — CSE 55 Batch
Key Courses:
- Artificial Intelligence
- Data Mining & Machine Learning
- Natural Language Processing
- Statistics & Probability
- Operating Systems
- Data Structures and Algorithms
- Computer Networks
- Database Management Systems
- Software Engineering
Thesis:
A Novel Interactive AI-Based Tool for Detecting Mental Stability Through Analysis of Human Voice
The undergraduate thesis experiments reported ~96% accuracy on their study setup. The later peer-reviewed Journal of Voice work used a different evaluation design and reported ~91% accuracy with ~0.97 ROC-AUC.
Supervisor: Dr. Md. Taimur Ahad (Associate Professor, Associate Head, Dept. of CSE; Coordinator, 4IR Research Cell, DIU)