Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I'm Mehdi Hajoub, an AI engineer specializing in production ML systems and model optimization. I design and deploy neural networks for medical applications, delivering measurable improvements in accuracy, latency, and reliability. I thrive at the intersection of research and production, with a strong foundation in MLOps, computer vision, and clinical AI deployment. I enjoy collaborating with clinicians and researchers across EPFL, CHUV, and Campus Biotech to turn ideas into robust, scalable solutions.…I'm Mehdi Hajoub, an AI engineer specializing in production ML systems and model optimization. I design and deploy neural networks for medical applications, delivering measurable improvements in accuracy, latency, and reliability. I thrive at the intersection of research and production, with a strong foundation in MLOps, computer vision, and clinical AI deployment. I enjoy collaborating with clinicians and researchers across EPFL, CHUV, and Campus Biotech to turn ideas into robust, scalable solutions.

Mehdi Hajoub

AI Engineer, AI Developer, AI Prompt Engineer, +7





I'm Mehdi Hajoub, an AI engineer specializing in production ML systems and model optimization. I design and deploy neural networks for medical applications, delivering measurable improvements in accuracy, latency, and reliability. I thrive at the intersection of research and production, with a strong foundation in MLOps, computer vision, and clinical AI deployment. I enjoy collaborating with clinicians and researchers across EPFL, CHUV, and Campus Biotech to turn ideas into robust, scalable solutions.…I'm Mehdi Hajoub, an AI engineer specializing in production ML systems and model optimization. I design and deploy neural networks for medical applications, delivering measurable improvements in accuracy, latency, and reliability. I thrive at the intersection of research and production, with a strong foundation in MLOps, computer vision, and clinical AI deployment. I enjoy collaborating with clinicians and researchers across EPFL, CHUV, and Campus Biotech to turn ideas into robust, scalable solutions.

Available to hire

I’m Mehdi Hajoub, an AI engineer specializing in production ML systems and model optimization. I design and deploy neural networks for medical applications, delivering measurable improvements in accuracy, latency, and reliability.

I thrive at the intersection of research and production, with a strong foundation in MLOps, computer vision, and clinical AI deployment. I enjoy collaborating with clinicians and researchers across EPFL, CHUV, and Campus Biotech to turn ideas into robust, scalable solutions.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Intermediate

Language

French

Fluent

English

Fluent

German

Intermediate

Italian

Beginner

Work Experience

Machine Learning Engineer Intern at Campus Biotech

January 1, 2025 - November 24, 2025

Deployed graph neural network system for neurodegenerative disease classification (12 disorders, 600+ fMRI scans) serving 15+ researchers with 99.2% uptime and sub-2-second inference latency. Built automated retraining pipeline with Airflow and Docker, reducing update cycle from quarterly to weekly and preprocessing time from 72h to 4h through stability-based feature selection (60% storage reduction).

AI Project Developer at EPFL AI Center

May 1, 2024 - May 1, 2024

Developed real-time sign language recognition system (1K videos, 50K frames) using Temporal Transformer architecture with CTC decoder, achieving 91% accuracy at 30 FPS. Compressed model from 42M to 11M parameters via pruning and quantization while maintaining 88% accuracy with 3× faster inference, enabling mobile deployment.

Machine Learning Researcher at EPFL LTS5 Lab

January 1, 2024 - January 1, 2024

Trained medical imaging models (ResNet-50, EfficientNet-B3) for cervical cancer detection on 200-image dataset, achieving 93% accuracy with cross-hospital validation (Geneva ↔ Cameroon). Improved domain generalization by 18% and reduced critical false negatives by 22% through adversarial domain adaptation and hybrid supervision techniques.

Machine Learning Engineer Intern at Campus Biotech, Geneva

January 1, 2025 - November 25, 2025

Deployed graph neural network system for neurodegenerative disease classification (12 disorders, 600+ fMRI scans) serving 15+ researchers with 99.2% uptime and sub-2-second inference latency; built automated retraining pipeline with Airflow and Docker, reducing update cycle from quarterly to weekly and preprocessing time from 72h to 4h through stability-based feature selection (60% storage reduction).

Machine Learning Engineer Intern at NeuroRestore (CHUV-EPFL), Lausanne

March 1, 2024 - March 1, 2024

Built multimodal NLP system to objectively quantify rehabilitation progress in spinal cord injury patients: automated transcription of 40+ hours of therapy sessions with medical language embeddings, synchronized to neuro-stimulation parameters and EMG data to measure movement quality and correlate verbal commands with motor recovery; optimized production inference by 65% (8.2s → 2.9s) through INT8 quantization and batch processing, enabling near-real-time clinical feedback; maintained deployment with Docker containers and CI/CD pipeline (GitHub Actions).

AI Project Developer at EPFL AI Center, Lausanne

May 1, 2024 - May 1, 2024

Developed real-time sign language recognition system (1K videos, 50K frames) using Temporal Transformer architecture with CTC decoder, achieving 91% accuracy at 30 FPS; compressed model from 42M to 11M parameters via pruning and quantization while maintaining 88% accuracy with 3× faster inference, enabling mobile deployment.

Machine Learning Researcher at EPFL LTS5 Lab, Lausanne

January 1, 2024 - January 1, 2024

Trained medical imaging models (ResNet-50, EfficientNet-B3) for cervical cancer detection on a 200-image dataset, achieving 93% accuracy with cross-hospital validation (Geneva ↔ Cameroon); improved domain generalization by 18% and reduced critical false negatives by 22% through adversarial domain adaptation and hybrid supervision techniques.