Skip to content
Mazen — sa/acc

Me

AI research & engineering.

I'm Mazen — born and raised in Dhahran, now in Riyadh. I build AI systems and the infrastructure they run on. Right now that's research and engineering at a government entity; before, the founding team at DEEP.SA and ML systems at Malaa. My academic research was real-time multi-object tracking and object detection, in the Deep Machine Vision lab at Oregon State. What I work on now is making agents reliable enough to put in front of real users — the harnesses, evals, and retrieval that decide whether they hold up.

Now
AgentsHarnessesEvalsLLM reasoningRetrievalEmbeddingsVLMs

Papers and projects are below. Essays are in Writing and event notes are in Talks.

On this page

Experience

Research

Papers and projects.

My academic work was real-time multi-object tracking and object detection. Now I work on making agents reliable enough to put in front of real users — most of that ships rather than gets published. The writing is on the posts page.

Publications

arXiv 2022
Masader Plus: A New Interface for Exploring +500 Arabic NLP Datasets
A web interface for browsing the Masader catalogue of Arabic NLP datasets — exploration, filtering, and an API over the backend.
arXiv
CVPR 2021
Discriminative Appearance Modeling with Multi-track Pooling for Real-time Multi-object Tracking
Most trackers model each target in isolation, which breaks down when objects look alike. We introduce a multi-track pooling module that updates appearance memory across all tracks at once, with a matching training strategy that generates hard tracking episodes online — real-time state-of-the-art on public MOT benchmarks under a bilinear-LSTM framework.
arXivCode
IBM Think 2019
Leveraging system architecture to optimize deep learning and AI-based processing of massive data sets
GPU expansion cards hit real limits once they have to move large data across the system bus. We evaluated several GPU architectures on massive audio, video, and DNA sequence workloads at Oregon State's Center for Genome Research and Biocomputing (CGRB), and showed where CPU–GPU integration is the actual bottleneck.
PaperSlides

Projects

2022
Detecting Anomalies in Object Appearance and Motion Dynamics
Oregon State University · DARPA Machine Common Sense
On the Passive Violation of Expectation team under Dr. Alan Fern. Built a tracker on Bilinear-LSTM and a Siamese network, a role-assignment module to decide which objects matter, and a rule-based reasoning agent that judges whether a scene is plausible. The team took 31 of 32 tracks in the final DARPA evaluation, April 2023.
PyTorch · Bilinear-LSTM · Siamese networks
2022
pyreqpp
Automatically updates missing module versions in requirements.txt to the latest.
Python
2021
onlysudo
A Discord and Twitch bot running style transfer, English and Arabic text completion, Arabic poetry generation, English QA, and TTS — built for Twitch streamers.
Python · Hugging Face · PyTorch · ONNX · TensorFlow · Node.js · Socket.IO
2021
Improving a Generative Visual Dialog model
Oregon State University
Took the model from 23.01% to 41.57% by swapping the image embedding for pre-trained CLIP and the word embedding for GloVe. Built a web application on top of it.
CLIP · GloVe · PyTorch
2020
How long was the poultry sick?
Oregon State University
An ML pipeline that auto-labelled over 100,000 bird sound samples using hidden Markov models, FFT, k-means and bootstrapping, then a deep learning classifier to detect illness from the calls — deployed to an NVIDIA Jetson Nano with TensorRT.
Keras · TensorFlow · SoX · TensorRT · Jetson Nano
2020
Labeling tool
Oregon State University
An image labelling tool that took throughput from 300 to 3,000 images per week. Used across multiple labs, including fieldwork in Africa with no internet access.
jQuery · Node.js
2020
Measuring interaction in sport
Oregon State University
A social network analysis system for understanding how players interact during a match.
C++ · Python · Cython · TensorFlow · Google Research Football
2019
Pedestrian Tracking and Privacy Preservation
Oregon State University · Senior Design
A computer vision system for the City of Portland that detects and obfuscates pedestrians' faces in real time, then tracks movement to study pedestrian and traffic behaviour for road safety.
OpenCV · PyTorch · Keras · Flask · Node.js · MongoDB
2019
Tech Data AI Demo
Oregon State University · CGRB
A site that races different hardware against each other running the lab's deep learning models. Sponsored by Tech Data, IBM, NVIDIA and OpenPower, and featured at IBM Think 2019.
Bootstrap · JavaScript · Node.js · NGINX · Bash
2019
Owl Sound Classification
Oregon State University · CGRB
Deep learning models that classify owl species from audio.
Keras · TensorFlow · OpenCV · NumPy · SoX
2019
Plankton Detection
Oregon State University · CGRB
Deep learning models for detecting plankton in imagery data.
Keras · TensorFlow · OpenCV · NumPy
2018
Image Captioning
A CNN–RNN model that generates captions from images: a convolutional network for feature extraction, an LSTM for generating the caption.
PyTorch · NumPy · OpenCV
2018
Self-Driving RC Car
A web app streaming live video from a Raspberry Pi-controlled RC car, plus the intelligent agent structure and model behind it.
Python · Node.js · Raspberry Pi
2017
Aces Up
Oregon State University
A card game built with a team. Won Best Web Application for Software Engineering I.
Java (Ninja) · JavaScript · Heroku
2016
Personal website
Oregon State University
An early version of this site, plus an NGINX config that redirects HTTP to HTTPS and maps networks.
Node.js · JavaScript · NGINX
2016
AI agent for a 2D grid game
Oregon State University
Hunt the Wumpus with a dynamic grid size, and an embedded agent that solves it using probability distribution methods.
C++

Education

M.Sc.
Oregon State University
Artificial Intelligence
Deep Machine Vision lab (Dr. Fuxin Li) — real-time multi-object tracking and object detection. Research on DARPA Machine Common Sense with the Passive Violation of Expectation team under Dr. Alan Fern.
B.Sc.
Oregon State University
Computer Science (AI)
Minor in Actuarial Science; VP of the OSU ML/AI Club. Earlier GPU and deep learning work at the Center for Genome Research and Biocomputing (CGRB).

Certifications

Community

Praise

I miss you already, Mazen. You had a huge impact on project success and are a delightful collaborator. It was a pleasure and a privilege working with you.
Jed Irvine Research Engineer, Oregon State University — on leaving the research group Jun 2023 ↗
I can say without hesitation that Mazen is the best TA I've worked with thus far. I had a lot of trouble this quarter with project 3 and reached out to another TA for help. After 45 minutes of typing back and forth, I was nowhere closer to understanding the project. I reached out to Mazen, who immediately hopped on a call for a debugging session. In 15 minutes, we resolved issues I had been stuck on for one week. He's gifted intellectually, but he's also very kind and didn't once blame me for not understanding something. He created a very open learning environment where I felt free to express my concerns and questions.
Ashley Owens Then a CS 372 student, now SWE II at Amazon — in an email to the faculty Jun 2021

One of more than ten students who wrote in with the same thing.

Thanks Mazen, you did a great job on the demo and really represented us well. Your help was crucial to this success.
Christopher M. Sullivan Director of the CGRB, Oregon State University — on the IBM Think 2019 demo Feb 2019 ↗

Alan Fern and Fuxin Li, the Oregon State professors who supervised the Machine Common Sense work, put the April 2023 result down to that approach — 31 of 32 tracks in the final DARPA evaluation — crediting the insistence on grounding the research in real software practice and evaluating models systematically rather than anecdotally. And across every group I've worked in, the feedback lands in roughly the same place: hard-working, and good company while doing it.

Life goals

Things I want to do before I die, borrowed as a format from Chip Huyen's List 100. If you have a recommendation, tell me.

Research & academia 2/12
  • 1 Get a Master's degree Sep 2022 — done
  • 2 Become one of the authors of a paper at a top-tier conference CVPR 2021 Jan 2021 — done
  • 3 Become the first author of a paper at a top-tier conference — not yet
  • 4 Get a PhD paused — paused
  • 5 Reach 500 citations — not yet
  • 6 Review for a top-tier conference — not yet
  • 7 Serve on a program committee — not yet
  • 8 Reproduce a landmark paper end to end — not yet
  • 9 Supervise someone to their first publication — not yet
  • 10 Teach a course — not yet
  • 11 Teach a graduate-level course — not yet
  • 12 Work as a scientist at one of the FAANG companies — not yet
Build & publish 2/8
  • 13 Publish a Python package pyreqpp Jul 2022 — done
  • 14 Get an affiliation on Twitch Jun 2020 — done
  • 15 Publish an NPM package — not yet
  • 16 Publish more than 50 posts 13 / 50 — not yet
  • 17 Write a book — not yet
  • 18 Get a patent — not yet
  • 19 Speak at an international conference — not yet
  • 20 Program or write every day for a year — not yet
Systems & scale 0/7
  • 21 Pre-train a model from scratch at real scale — not yet
  • 22 Train a model on 1,000+ GPUs — not yet
  • 23 Write a CUDA kernel that beats the library implementation — not yet
  • 24 Implement a transformer from scratch with no references — not yet
  • 25 Write a training run that recovers from a crash unattended — not yet
  • 26 Profile and fix a real GPU bottleneck — not yet
  • 27 Train something on a budget everyone said was impossible — not yet
Agents & harnesses 0/8
  • 28 Build an agent that runs autonomously for a week — not yet
  • 29 Ship an eval harness other people adopt — not yet
  • 30 Get an MCP server into wide use — not yet
  • 31 Have an agent open a pull request a maintainer merges unedited — not yet
  • 32 Run agents in production against real user traffic — not yet
  • 33 Write an agent framework from scratch — not yet
  • 34 Top a public agent benchmark — not yet
  • 35 Build a memory system that beats RAG baselines — not yet
Libraries to contribute to 0/6
Open source & reach 0/6
  • 42 Get 1,000 GitHub stars on a project — not yet
  • 43 Build a tool used by 10,000 people — not yet
  • 44 Maintain a project with outside contributors — not yet
  • 45 Have my work cited in someone else's paper — not yet
  • 46 Have a stranger tell me they use something I built — not yet
  • 47 Ship something that runs in production for five years — not yet
Found & lead 1/3
  • 48 Start a non-profit organization sa/acc Dec 2024 — done
  • 49 Start a startup — not yet
  • 50 Mentor 100 Saudi engineers 47 and counting — not yet
Money & independence 0/3
  • 51 Reach financial independence — not yet
  • 52 Make real income from something that isn't a salary — not yet
  • 53 Fund someone else's work — not yet
Craft & mind 0/5
  • 54 Learn an instrument — not yet
  • 55 Master a new hard skill in under 48 hours — not yet
  • 56 Write down how I learn things fast — not yet
  • 57 Teach someone else the 48-hour method — not yet
  • 58 Build a piece of furniture — not yet
Legacy 0/3
  • 59 Have someone credit my work as why they got into AI — not yet
  • 60 Leave a project that outlives my involvement — not yet
  • 61 Write something still read in ten years — not yet
Places 1/2
  • 62 Live in another country (US) — done
  • 63 Live in a different country for over 2 years (other than the US) — not yet
Chinese 0/5
Travel & place 0/10
  • 69 Visit 50 countries — not yet
  • 70 Set foot on every continent — not yet
  • 71 See the northern lights — not yet
  • 72 See a total solar eclipse — not yet
  • 73 Dive the Red Sea — not yet
  • 74 Cross the Empty Quarter — not yet
  • 75 Sleep under the stars in AlUla — not yet
  • 76 Hike to Everest base camp — not yet
  • 77 Ride in a hot-air balloon — not yet
  • 78 Take a multi-day train journey — not yet
Table & craft 0/3
  • 79 Eat sushi in Japan — not yet
  • 80 Learn to make bread properly — not yet
  • 81 Grow something I can eat — not yet
Always 0/2
  • 82 Build and ship in public ongoing — ongoing
  • 83 Leave people better than I found them ongoing — ongoing

Timeline

Milestones

The individual notes behind the roles, moves, and community milestones.

May 2026
Left Cursor
Wrapped up a year as Cursor Ambassador for Riyadh and Saudi Arabia.
Apr 2026
Joined Confidential Government
Building AI systems at the core, in pursuit of sa/acc.
Mar 2026
Founded sa/acc
Started sa/acc — Saudi Acceleration, building and shipping in public.
Feb 2026
Left DEEP.SA
Wrapped up as AI Staff Engineer on the founding team.
Dec 2025
Joined DAL | دال
Joined DAL, a non-profit association supported by Misk Foundation & SDAIA.
Jun 2025
Joined DEEP.SA
Joined DEEP.SA as AI Staff Engineer on the founding team.
May 2025
Last Day @ Malaa
Two years at Malaa Technologies — from ML engineer to production systems.
May 2025
Became a Cursor Ambassador
Joined Cursor as an ambassador for Riyadh and Saudi Arabia.
Oct 2023
PyTorch Ambassador Award 2023
Recognized by PyTorch for community contributions and advancing deep learning education in the MENA region.
May 2023
Joined Malaa Technologies
Started as Senior Machine Learning Engineer at Malaa, a Saudi fintech.

Elsewhere

Find me on X, LinkedIn, and GitHub, or email hello@ma7.dev.