Role overview
As a Machine Learning Engineer on our Foundational team in Paris, you will build the "brain" of our tactical robots. You will design and scale large-scale, multi-modal foundational models that learn robust representations of the battlefield using Self-Supervised Learning (SSL) from massive amounts of unlabelled Electro-Optical (EO) and Infrared (IR) data. Your work provides the critical foundational weights that our Edge AI team distills into hyper-accurate models running on tactical hardware.
What you'll work on
- Multi-Modal SSL Architecture Design: Design neural network architectures (Vision Transformers) and loss functions (Masked Autoencoders, Contrastive Learning) to jointly learn from paired and unpaired EO and IR data.
- Distributed Training Infrastructure: Manage and optimise training pipelines across multi-node GPU clusters, handling mixed-precision training and data loading.
- Representation Evaluation: Develop metrics and linear-probing benchmarks to prove the latent space captures useful semantic features before distillation.
- Data Strategy: Audit existing EO/IR data lakes and implement cross-attention mechanisms to fuse diverse sensor features.
- Cross-Functional Collaboration: Sync with Data Engineers on ingestion pipelines and collaborate with the Edge AI team to ensure high-performance model handoffs.
What we're looking for
- Educational Background: A PhD or a highly research-focused MS in Computer Science, Machine Learning, Computer Vision, or Applied Mathematics.
- Proven Experience: * Minimum of 5-6 years of experience for senior levels.
- Experience training and scaling deep learning vision models (ViTs, CNNs) from scratch in multi-GPU/multi-node environments.
- Successful application of novel SSL or multi-modal architectures (e.g., CLIP, MAE, DINO) to real-world, non-standard imaging data (IR, SAR, or hyperspectral).
- Technical Proficiency: * Hardcore PyTorch engineering skills combined with deep mathematical intuition for representation learning
- Knowledge of system-level languages (C++, Rust, or Go) and resource optimisation for edge computing.
- Complexity & Leadership: Ability to architect state machines for fault-tolerant data pipelines and mediate technical trade-offs between hardware and algorithm teams.
- Commitment & Mindset: 100% dedication to Harmattan AI’s mission of providing an ethical defence edge to allied countries. A hybrid researcher-engineer mindset that treats data quality as seriously as algorithm design
We look forward to hearing how you can help shape the future of autonomous defense systems at Harmattan AI.
Tags & focus areas
Used for matching and alerts on DevFound Fulltime Ai Machine Learning Robotics