Open to project work and research collaborations
Hello, I'm

Maidul Hasan Masum

I'm an AI engineer who works on computer vision. At Graymatics in Singapore I build video analytics that runs on live camera feeds. In research, I work on small, efficient detection models for devices with very little memory and power.

About

Research that has to work on real cameras

I like work where an idea has to survive contact with the real world. At Graymatics I build and ship video analytics: counting and classifying vehicles, checking safety and PPE compliance, spotting shoplifting, estimating the weight of sorted waste, and summarising long recordings so a reviewer can find the incident quickly. These run on live cameras, so latency and reliability matter as much as accuracy.

My research asks the same question from the other side: how small can a model get before it stops being useful? My MSc looks at object detection on tiny wireless sensor devices, and I try to measure accuracy together with what it costs to run.

I also teach. From 2023 to 2026 I tutored English at EF English Academy, and I have been a teaching assistant at university. Explaining an idea clearly is the fastest way I know to find out whether I really understand it.

3+ yearsbuilding production computer vision
16pull requests merged into 8 open-source vision projects
4papers published, with two more accepted
IEEESensors Journal, first-author paper
Experience

Where I've worked

Production computer vision first, with earlier roles in machine learning for banking and software.

Sep 2023 to present

AI Algorithm Engineer

Graymatics, Singapore

I design, train and deploy the models behind video analytics products across five commercial areas: waste, traffic, retail, safety and security. The systems serve clients in Singapore and in international markets.

  • Waste and weight. Classify sorted waste and estimate its weight by calibrating pixels to real units, with automatic reports.
  • Traffic. Detect and classify vehicles, count them by lane and region, measure queues and congestion, and estimate speed.
  • Retail. Detect shoplifting, count people, analyse footfall and groups, and blur faces to protect privacy.
  • Safety and compliance. Check hairnets and PPE, verify ID cards, and flag weapons, violence and unsafe kitchen practices.
  • Tracking and summaries. Multi-object tracking with SORT and ByteTrack, behaviour analysis over time, and video summaries that speed up incident review.
  • Running it for real. Low-latency pipelines with post-processing, QA, dashboards and alerts, deployed in containers on AWS.
PyTorchOpenCVCUDATensorRTONNXTritonDockerAWS
Sep 2022 to Feb 2023

AI Research and Development Intern

Maybank, Johor Bahru, Malaysia
  • Tuned OCR extraction for faster processing and better accuracy on documents.
  • Updated the chatbot's content and features to improve how users engaged with it.
  • Developed and fine-tuned machine learning and NLP models to improve predictions.
Jan to Aug 2022

AI Development Intern

proceedit Trading (Expertel S.A.), Barcelona, working remotely

Built and optimised machine learning applications, handling data analysis and preprocessing. Worked with an international team using AWS EC2 and Jira.

Oct to Dec 2020

Software Engineering Virtual Internship

JPMorgan Chase and Co, on Forage

Completed JPMorgan Chase's software engineering job simulation and ranked in the top 2% of all Forage students. I connected to a stock price data feed, worked with JPMorgan Chase frameworks and tools, and displayed the data visually for traders.

Education

Oct 2024 to Dec 2026

Master of Computer Science (by research)

Universiti Malaya, Malaysia

Fully funded as a Graduate Research Assistant. My research area is edge AI and wireless sensor networks, with agricultural computer vision as the main application. My dissertation is about a small object detector for wireless vision sensors. The thesis is submitted and waiting for examination.

Sep 2019 to Aug 2023

Bachelor of Computer Science (Software Engineering)

Universiti Teknologi Malaysia, Johor Bahru

First Class Honours with a CGPA of 3.70 out of 4.00. I received the Dean's Award and the Best Undergraduate Research Award. My thesis compared CNN, DCNN, Fast and Faster R-CNN, YOLO and transfer learning for facial expression recognition on the CK+, JAFFE and FER2013 datasets.

Research

Efficient models for small devices

I try to treat cost as part of the design from the start, and to compare methods fairly: same data, same conditions, accuracy always reported next to compute.

LightYolo-SSO

Also called WSN-YOLO

An object detector for wireless sensor nodes that have very little memory and power. Instead of shrinking an existing model, I searched for the architecture directly using Sperm Swarm Optimization, scoring each candidate on accuracy, FLOPs, size and speed.

  • Searches the architecture directly instead of shrinking an existing model
  • Scores every candidate on accuracy together with FLOPs, size and speed
  • Compared fairly with compact YOLO baselines on the same sensor dataset

Results are being updated as the work continues.

View the code on GitHub

Adaptive LightYolo-SSO

Early-exit inference

Most inputs are easy, so running the whole network on them wastes energy. I added two early exits to the LightYolo-SSO backbone. Easy samples leave early, and only the hard ones go all the way through.

  • Two small exit branches on the LightYolo-SSO backbone
  • Confident samples leave early and skip most of the computation
  • Savings are measured in operations, with hardware tests still to come

Results are being updated as the work continues.

View the code on GitHub

Research roles

Oct 2024 to present

Graduate Research Assistant

Universiti Malaya

I work on moving computer vision and NLP methods into systems that can actually be deployed, run reproducible experiments, and write up what the results mean in practice. I co-author the resulting papers.

Aug 2020 to Sep 2023

Research Assistant

Universiti Teknologi Malaysia

Worked with faculty on AI and IoT research, including a study of how universities adopt IoT for e-learning, which was later published in Frontiers in Psychology. Also helped on an automated wheelchair and a Raspberry Pi race timer.

Aug 2023

Committee Member, Technical Team

4th International Conference on Applied Computing

I was on the organising committee for this hybrid conference. I helped run the technical side, coordinated parts of the event, and supported promotion and attendee outreach.

Work with me

Computer vision projects for small teams

Outside my job and research I take a small number of projects for small teams and agencies. These are the jobs I take most often. Prices are on request.

YOLO fine-tune and TensorRT export

About 5 working days

A detector trained for your objects and exported to run fast on your GPU or edge device. You get the weights, the ONNX and TensorRT files, an accuracy report and measured latency.

Latency and accuracy audit

About 3 working days

I go through your detection pipeline and find where the time and the accuracy are lost. You get a written report with a ranked list of fixes.

Dataset and evaluation check

About 2 working days

I check your labels, format conversions and metric code for the errors that make accuracy numbers wrong. You get a report with a fix for each one.

Custom model or full pipeline

Scoped after a call

Need a custom model, data labelling, or a full camera pipeline with alerts? Tell me the problem and I will come back with a scope and a timeline. For larger jobs I can put together and manage a small team.

How I work

  • It starts with an email or a short call about the problem. I send back a fixed scope, timeline and price before any work begins.
  • Every speed or accuracy figure comes from a run you can repeat on your own hardware.
  • You own the code, the weights and the report.
  • I work in English from Kuala Lumpur (UTC+8), with a few hours of overlap with Europe and Australia each day.

You can try my work first on the CV Test Bench: run a detector on your webcam or add your own model.

To start, email mhmasum6875@gmail.com and tell me what you are trying to do.

Skills

What I work with

Tools I use every week to take a model from an idea to a stream running in production.

AI and machine learning

PyTorchTensorFlowscikit-learnOpenCVHugging FaceUltralytics YOLO (v5 to v11)

Computer vision

Object detectionImage classificationInstance segmentationTracking (SORT, ByteTrack, DeepSORT)Behaviour analytics

Deployment

TensorRTONNXTritonDockerAWS EC2CUDAReal-time inference

Language and LLM work

Prompt engineeringText classificationSummarisationInformation extractionOCRTransformer fine-tuning

Programming

PythonC++JavaSwiftBashSQLDartMATLAB

Web

React.jsNode.jsExpress.jsMySQLHTMLCSS
Publications

Papers

My own work plus collaborations. My name is in italics. The full list is on Google Scholar.

2025
Toward Efficient and Accurate Anomaly Detection in Wireless Sensor Networks: A Review First author
M. H. Masum, B. Raj, S. S. Showmik. IEEE Sensors Journal, vol. 25, no. 24, pp. 43585-43602. DOI, IEEE Xplore
2024
Critical Review on Resource Scheduling in IaaS Clouds: Taxonomy, Issues, Challenges, and Future Directions
S. H. H. Madni, M. Faheem, M. Younas, M. H. Masum, S. Shah. The Journal of Engineering (IET and Wiley), vol. 2024, no. 8, e12420. DOI
2024
Underwriting in the Age of Insurtech: A Comparative Study of Algorithms and Their Efficacy First author
M. H. Masum, R. Hassan, H. Madni, M. A. Ahsan, I. M. Rahman. Proceedings of the 6th UTM Undergraduate Research Conference (URC 2023), pp. 317-323. ResearchGate
2022
Factors Influencing the Adoption of IoT for E-Learning in Higher Educational Institutes in Developing Countries
S. H. H. Madni, J. Ali, H. A. Husnain, M. H. Masum, S. Mustafa, J. Shuja, M. Maray, S. Hosseini. Frontiers in Psychology, vol. 13, article 915596. DOI
Soon
Edge AI-Driven Wireless Vision Sensor Networks for Collaborative Robotic Perception Accepted First author
M. H. Masum, B. Raj, S. S. Showmik, M. S. Hossain, A. Oza. INTRAC 2025, Springer, in press.
Soon
Trustworthy and Physics-Aware Deep Learning for Myocardial Motion Estimation: Principles, Methods, and Challenges Accepted
S. S. Showmik, M. Y. I. Idris, Z. Yu, M. H. Masum. INTRAC 2025, Springer, in press.
Review
Test-Time Scaling in Computer Vision: A Unified Review of Inference-Time Compute for Visual Understanding, Generation, and Real-Time Deployment Under review First author
M. H. Masum, Z. H. Ahmed, A. M. Zain, M. M. R. Masud. Neural Computing and Applications, Springer.
Review
Bridging Accuracy and Deployability: A Cross-Domain Systematic Review of Encoder Transformers for Text Classification Under review First author
M. H. Masum, S. S. Showmik, A. M. Zain, M. I. Solihin, T. Tibyani, A. Nafis, L. Hasan. Knowledge and Information Systems, Springer.
Review
Deep Learning for Facial Expression Recognition: A Survey of Classical, Transformer, Multimodal and Generative Approaches Under review First author
M. H. Masum, R. Hassan, A. M. Zain, T. Tibyani, S. H. Madni, S. S. Showmik, M. T. Aumanee. The Visual Computer, Springer.
Awards

Recognition

Research, academic and competition awards from my undergraduate years and since.

2023

Best Undergraduate Research Award

UTM Undergraduate Research Conference

2023

Dean's Award

Faculty of Computing, UTM

2023

Second prize, ACF2023 Idea Contest

ASEAN Career Fair with Japan, for an AI application concept

2022

Third prize, Youth Regional Affairs Dialogue

NTU, Singapore, for a project on the future of AI

2021

First runner-up, Sayembara Bahasa Melayu

Malay poetry competition, international student category

Other recognition

Best Speaker, Robokar runner-up, and second runner-up at the Product Positioning and Branding Workshop and at Designing Doodle

Projects

Things I've built

Tools and smaller projects from outside my main research.

CV Test Bench

Web app for testing vision models

A dashboard for testing a vision model on live video. Add RTSP streams or video files, attach a model with a zone and an alert rule, and watch the annotated feed. The hosted demo runs in your browser, with open YOLO models and a lab of eight vision tasks.

PythonFastAPIOpenCVYOLOJavaScript
Open the live demo Try the CV Lab (8 vision tasks) View the code on GitHub

Vision Field Notes

Blog

Practical articles on computer vision and edge AI: training detectors, choosing hardware, running models on small devices, and keeping video analytics reliable on real cameras.

YOLOEdge AIVideo analytics
Read the blog

MetaTr@cker

Android app

A free fitness tracker with no ads, made during the COVID-19 lockdowns when gyms were closed. It has BMI and FFMI calculators, tracking for weight, body fat, muscle and measurements, and custom workout plans.

JavaFirebaseRetrofit
View the code on GitHub

Bluetooth Robot Controller

Flutter app and Arduino

A phone app that connects to an Arduino over Bluetooth to drive a robot or RC car and show what the board sends back. The repository includes the Arduino sketch that turns each command into motor control.

FlutterDartArduino
View the code on GitHub
Beyond work

Leadership and teaching

I enjoy explaining things and organising people, and I've done both throughout university.

Teaching

  • English Language Tutor, part-time, Jun 2023 to Aug 2026EF English Academy. I taught IELTS, PET and Cambridge English programmes to students at different levels, in person and online.
  • Teaching AssistantUTM-IET Capstone Experience 5.0
  • Academic support volunteer, June 2022Answered student questions and helped run practical and tutorial sessions at UTM.

Leadership

  • Head of Business, Entrepreneurship and Networking, 2022 to 2023Kolej Dato' Onn Jaafar, UTM. Led networking and student initiatives with university and outside partners.
  • Vice President, 2021 to 2022Kolej Datin Seri Endon, UTM
  • Secretary, 2021 to 2022International Student Society Central, UTM
  • Head of Department Secretary, 2020 to 2022UTM Global Buddies, which welcomes international students
  • Delegate reception, May 202313th UTM University Presidents Forum

Scholarships and summer programmes

  • ICP Scholarship, 2023Fully funded International Cultural Program at Hasanuddin University, Indonesia
  • ITS-UTM Leadership Programme, 2023Selected participant, Surabaya, Indonesia
  • Youth Regional Affairs Dialogue, 2022Fully funded, NTU University Scholars Programme, Singapore
  • UNPAD International Virtual Summer Programme, 2021Marine tourism and coastal vulnerability
  • UKM U-LEAD, 2021Virtual Global Leadership Programme

Let's talk

I'm open to fixed-scope project work and research collaborations. I'm especially curious about video analytics for ports and shipping: vessel detection and tracking, reading hull and IMO numbers, berth occupancy, underwater inspection, and spotting spills. If you work in that area and want to try something, get in touch.

↑