AI Research Intern

Description

Harmattan AI is building autonomous and scalable defense systems. They are expanding their teams and capabilities to deliver mission-critical systems to allied forces.

Responsibilities

• Design and implementation of deep learning models for computer vision tasks.
• Research and experimentation with CNNs and Vision Transformers.
• Model compression techniques such as knowledge distillation.
• Quantisation-aware training (QAT) and post-training quantisation (PTQ).
• Network and dataset pruning strategies.
• Design of efficient architectures for edge and embedded systems.
• Dataset curation, balancing, and bias mitigation.
• Experimental design, ablation studies, and reproducibility practices.
• Robust evaluation using appropriate metrics (e.g., mAP, IoU, calibration).
• Failure case analysis and robustness testing under distribution shifts.

You will be encouraged to read, analyse, and implement ideas from leading conferences such as CVPR, ICCV, ICLR, and NeurIPS.

Requirements

• Technical Background: Solid foundation in Deep Learning (preferably PyTorch).
• Experience with CNNs and/or Transformers (academic or project-based).
• Understanding of bias–variance trade-offs and generalisation.
• Familiarity with optimisation fundamentals and basic probability.
• Experience or strong interest in model compression techniques.
• Interest in hardware-aware and efficient model design.

Also Nice to Have

• Experience with ONNX, TensorRT, TFLite or LiteRT.
• Familiarity with experiment tracking tools (e.g., W&B, MLflow).
• Experience conducting ablation studies.
• Exposure to dataset curation or annotation processes.
• Prior participation in research projects or conference work.

What You Will Gain

• Hands-on research experience in applied computer vision.
• Exposure to model optimisation and edge deployment challenges.
• Experience reading, implementing, and evaluating cutting-edge research.
• Mentorship from experienced researchers and engineers.
• Opportunity to contribute to publications or conference submissions.