machine learning engineer for AI agents

Digital Waffle is building an AI agent that handles genuine tasks for everyday users, including running errands, managing workflows, and maintaining context across long and complex conversations. The product focuses on reliable machine learning systems that operate at scale.

• Build end-to-end pipelines across data, training, evaluation, and inference;
• Adapt and fine-tune models using LoRA, QLoRA, SFT, DPO, and distillation;
• Architect inference systems that meet real latency and cost constraints;
• Create data pipelines for high-quality synthetic and real-world training data;
• Run evaluations covering robustness, safety, bias, and production behaviour;
• Own deployment, including GPU optimisation, quantisation, memory efficiency, and scaling;
• Work directly with application engineers to integrate ML into backend, mobile, and desktop products;
• Take ownership of ambiguous problems from zero to one;
• Ship, iterate, and learn from production.

• Deep understanding of deep learning and transformer architectures;
• Proven experience training, fine-tuning, or shipping large-scale models in production;
• Strong knowledge of at least one major ML framework, such as PyTorch or JAX;
• Familiarity with distributed training and inference tooling, including DeepSpeed, FSDP, Megatron, ZeRO, or Ray;
• Engineering discipline and ability to write readable, robust, and maintainable code;
• Experience optimising for GPU constraints, including quantisation, mixed precision, and memory;
• Comfortable taking ownership of ambiguous problems from zero to one;
• Nice to have: LLM inference frameworks such as vLLM, TensorRT-LLM, or FasterTransformer, RLHF methods including PPO, DPO, or ORPO, open-source contributions to ML or systems libraries, scientific computing, compiler, or GPU kernel experience, multimodal or diffusion model background, large-scale data processing with Arrow, Spark, or Ray.

• Remote role across the European Union;
• Work closely with research and engineering leadership;
• Real influence over architecture evolution and direct impact on users.