Available for ML / CV roles · Cairo, Egypt

Mohanad Ashraf Ramadan

Machine Learning Engineer

I build production-ready ML systems from raw, messy data. Specialized in Computer Vision, NLP, and Recommendation Systems — from problem formulation and modeling to scalable deployment and real-time monitoring

Mohanad Ashraf Ramadan, Machine Learning Engineer
Focus
Computer Vision
Best mAP@50
98.5%
Realtime
16 ms / frame
Degree
CS & Statistics
[ About ]

The person behind the models

Machine Learning Engineer with hands-on experience across the full breadth of ML.

Machine Learning Engineer with hands-on experience across the full breadth of ML: supervised and unsupervised learning, deep learning, computer vision, NLP, recommendation systems, and time series forecasting.

That range shows up in the work itself — from a custom computer vision pipeline for football tracking, to recommendation engines, classification systems, and NLP pipelines. Computer vision is where the most advanced systems live, but it's one specialty among several, not the whole story.

Currently pursuing a B.Sc. in Computer Science & Statistics at Helwan University (expected 2027), while training across three concurrent tracks: DEPI, EM Business Solutions, and Elevvo Pathways.

Skilled across the full ML lifecycle — data collection and annotation, preprocessing and feature engineering, model training and evaluation, and deployment via Streamlit and FastAPI. Comfortable independently owning a project from raw data to production-ready application, including designing custom algorithms for tracking and decision logic.

Cairo, EgyptB.Sc. CS & Statistics · Helwan Universitymohand12ashraf12@gmail.com
0+
ML projects shipped
0.0%
Best mAP@50 achieved
0
Real-time CV systems
0ms
Fastest inference time
[ Featured work ]

Systems built end-to-end

Hover a card to pull up the model card: accuracy, throughput and architecture.

AI-Powered Football Match Analysis preview
94.2%
Accuracy
38 FPS
Throughput
-71%
ID switches
model: YOLO + ByteTrack + Kalman

AI-Powered Football Match Analysis

End-to-end computer vision pipeline turning raw match footage into possession stats, event detection and homography-corrected heatmaps.

  • ▸Custom Kalman-filter ball tracker with dynamic gating and confirm-before-commit logic to reject false positives.
  • ▸KMeans jersey-colour clustering on top of ByteTrack for automatic team classification.
  • ▸Player ID stitching via Hungarian assignment and union-find chaining across occlusions and camera cuts.
  • ▸Two-pass architecture separating detection from rendering for accuracy and speed.
PythonYOLOOpenCVByteTrackStreamlit
View source
Productive vs. Distracted Detection preview
98.5%
mAP@50
16 ms (62 FPS)
Inference
99.8%
Precision
model: YOLO (benchmarked vs. RT-DETR)

Productive vs. Distracted Detection

A real-time engagement monitor built from scratch: custom recorded dataset, annotation, and a head-to-head benchmark of two detection architectures.

  • ▸Recorded and annotated a 900-image my own dataset across multiple camera angles, backgrounds and desk setups.
  • ▸Benchmarked YOLO vs. RT-DETR - YOLO chosen for ~2x faster inference at comparable accuracy.
  • ▸Rule-based alert engine with configurable timers to flag sustained distraction in real time.
  • ▸Detections cross-validated against mouse and keyboard activity to suppress false alerts.
PythonYOLORT-DETROpenCVStreamlit
View source
[ Technical expertise ]

The stack behind the systems

Full ML lifecycle ownership: data collection and annotation, training, evaluation, and deployment.

Programming & Tools

PythonSQL (MySQL, PostgreSQL)Git & GitHubJupyter NotebookVS CodeColab

Machine Learning & Deep Learning

Scikit-learnPyTorchTensorFlowSupervised & Unsupervised LearningBoosting AlgorithmsNeural NetworksCNNsHyperparameter TuningFeature EngineeringModel Evaluation & Optimization

Computer Vision

YOLO (v5–v11, YOLO26)RT-DETROpenCVObject DetectionMulti-Object Tracking (ByteTrack)Kalman FilteringImage ClassificationImage SegmentationTransfer Learning

NLP & Generative AI

Hugging Face TransformersFine-tuning Pre-trained ModelsRAG-based applicationsText Classification

Deployment & MLOps

StreamlitFastAPIDocker (exposure)Azure AI Fundamentals (exposure)

Data Analysis & Visualization

PandasNumPyMatplotlibSeabornPower BI (basic)
[ How I think ]

From understanding the problem to shipped model

The loop I run on every project, whether it's a tracking pipeline or a tabular classifier.

  1. 01

    Understand the problem

    Before a single line of training code: what decision does this model change, what does failure cost, and what does 'good enough' actually mean in production?

  2. 02

    Analyse the data

    EDA, class balance, leakage checks, annotation or label quality. When the dataset doesn't exist, I collect and annotate it myself, whether that's images, text, or tabular records.

  3. 03

    Build a baseline

    The simplest model that could work, wired end-to-end. A plain baseline that runs beats a clever one that never ships, and it sets the bar every iteration has to beat.

  4. 04

    Iterate & optimise

    Architecture benchmarks (classic ML vs. deep nets, CNNs vs. transformers), augmentation, tuning, and custom logic whenever off-the-shelf tools fall short.

  5. 05

    Deploy & evaluate

    Streamlit or FastAPI, latency and resource budgets measured on real hardware, and monitoring that tells me when reality drifts away from the validation set.

[ Journey ]

Where I've trained and built

Three concurrent ML tracks alongside a Computer Science & Statistics degree.

2026 - Present

AI & Data Science Track

EM Business Solutions Scholarship
  • ›Completed intensive project-based training across AI, Data Science and ML workflows.
  • ›Built ML and deep learning models with Scikit-learn and PyTorch, including feature engineering and model optimisation.
  • ›Developed NLP and Generative AI applications using Hugging Face and RAG-based approaches.
  • ›Applied computer vision: image processing, classification, object detection (YOLO), object tracking (ByteTrack) and Vision Transformers.
  • ›Deployed ML applications with Streamlit and FastAPI, with hands-on exposure to Docker.
2026

Machine Learning Intern

Elevvo Pathways
  • ›Built and optimised end-to-end ML and deep learning pipelines for classification, regression and clustering problems.
  • ›Applied core ML techniques with Scikit-learn and developed deep learning models with TensorFlow and Keras.
  • ›Conducted data preprocessing, feature engineering and model evaluation, including image resizing for computer vision tasks.
  • ›Developed and fine-tuned CNN architectures, leveraging transfer learning for high accuracy on image classification benchmarks.
2025 - Present

Machine Learning Engineer Track

Digital Egypt Pioneers Initiative (DEPI)
  • ›Completed intensive training covering the full AI lifecycle, from data preprocessing to model deployment.
  • ›Performed data preprocessing, feature engineering and EDA with Pandas, NumPy and Scikit-learn.
  • ›Implemented ML algorithms (regression, classification, clustering) and built deep learning models for predictive tasks using neural network architectures.
  • ›Applied NLP techniques for text analysis and computer vision methods for image-based tasks.
  • ›Gained hands-on experience deploying models with MLflow, MLOps practices and Microsoft Azure AI services.
Expected 2027

B.Sc. Computer Science & Statistics

Helwan University, Faculty of Science
  • ›Data structures & algorithms, database systems, machine learning & data mining.
  • ›Probability, distribution theory and order statistics underpinning applied ML.
[ Get in touch ]

Ready to build intelligent systems together?

Open to opportunities in machine learning, computer vision, NLP and end-to-end AI systems.

mohand12ashraf12@gmail.com +20 112 171 0632 Cairo, Egypt