Haldorix, a startup studio based in Casablanca, is on a mission to transform operational challenges into scalable business opportunities through technology. We identify recurring problems across manufacturing, retail, and logistics, and turn them into dedicated solutions that become standalone products. Our ventures include DeVisu, a computer vision solution for operational optimization, and Nitra3AI, a generative AI platform for industrial performance enhancement.

Position Overview

As a Computer Vision / Machine Learning Engineer – Model Training, you will play a pivotal role in building a real‑time productivity monitoring system for textile production lines. Your responsibilities will span the entire model lifecycle, from data analysis and pipeline design to deployment and continuous improvement.

Key Responsibilities

• Research & Discovery: Analyze existing datasets, identify optimal sampling and labeling strategies, and evaluate prior model performance to determine improvement paths through Active Learning.
• Design & Definition: Set up and structure video training pipelines for short clips, design and test data augmentation workflows for robustness, and fine‑tune the FACT‑tiny architecture on domain‑specific data.
• Execution & Collaboration: Implement GPU‑optimized training using mixed precision and TensorBoard, manage model checkpoints and retraining cycles during Active Learning, and work closely with the engineering team to integrate the model into the production system.
• Innovation & Learning: Explore alternative architectures such as SlowFast or 3D CNNs for performance comparison, document key findings and best practices for future model deployments.

Required Qualifications

• 2+ years of experience in deep learning model training and optimization.
• Strong proficiency with PyTorch.
• Experience with video architectures such as FACT, SlowFast, or 3D CNNs.
• Solid understanding of data augmentation tools (Albumentations, imgaug).
• Familiarity with GPU optimization, mixed precision, and training monitoring (TensorBoard).
• Proven ability to iterate quickly and improve model performance under time constraints.

Nice to Have

• Experience in industrial vision or manufacturing contexts.
• Exposure to MLOps workflows (versioning, checkpoints, pipeline reproducibility).
• Familiarity with Active Learning loops in computer vision.
• Hands‑on experience with real‑time inference pipelines or edge deployments.

Benefits

• Opportunity to work on a cutting‑edge applied AI project with real industrial impact.
• Close collaboration with an expert technical team and startup founders.
• Freedom to experiment, learn, and push the boundaries of real‑time vision systems.

Recruitment Process

1. Jobzyn AI interview (25–45 min).
2. Technical interview (1h) with the Lead Developer or Technical Architect.
3. Practical test (2–3h) based on a real‑world use‑case.
4. Final interview with the team and partners.

Join us to shape the future of industrial vision and make a tangible impact on production efficiency.