UK, Europe, USA
Recruit engineers who can design, train, adapt, evaluate and operate models, not just call an AI API.
Technical screening by developers and CTOs is included.
We recruit machine learning engineers, AI developers, LLM engineers and AI research engineers. The assessment goes beyond prompting and model APIs. We test whether candidates understand model architecture, data preparation, training dynamics, fine-tuning, evaluation, inference optimisation and the software needed to run models reliably in production.
When to hire AI developers: A team developing a custom or domain-specific model. An AI-native product fine-tuning and evaluating foundation models. A company moving trained models into reliable production inference.
We establish whether the role involves training a model from scratch, adapting a foundation model, building retrieval systems, improving inference, or owning the complete machine learning lifecycle.
We ask candidates to explain computation graphs, backpropagation, attention, tensor shapes and memory use. They should be able to select an architecture for the problem, justify that choice and describe how they would implement and debug the training loop rather than relying only on a high-level library call.
We probe dataset construction, tokenisation, loss functions, optimisers, learning-rate schedules, checkpoints, distributed training, overfitting and data leakage. Candidates compare full fine-tuning with parameter-efficient methods, explain LoRA rank and quantisation trade-offs, and design evaluation that detects regressions or catastrophic forgetting.
We ask candidates to build meaningful offline and online evaluations, separate retrieval failures from generation failures and reason about embedding and reranking quality. For serving, we test batching, caching, quantisation, GPU memory, latency, throughput, model versioning, drift monitoring and safe rollback.
AI engineers must be honest about what the evidence shows, resist overstating model capability and design experiments that can disprove their own assumptions. We value curiosity, responsible judgement and clear communication of uncertainty.
Candidates must show that they can create and improve models, understand the mathematics and systems beneath the frameworks, and make evidence-based decisions throughout the machine learning lifecycle.
Neural architectures, objectives, optimisation, gradient behaviour and the reasons a model learns or fails to learn.
Dataset quality, labelling, leakage prevention, reproducibility, baselines, ablations and statistically sound evaluation.
Full fine-tuning, PEFT, LoRA and QLoRA, preference optimisation, distillation and domain-specific model behaviour.
Model packaging, GPU utilisation, serving, observability, versioning, cost, latency and reliable production operation.
Tool names are not enough. Across every role, we look for technical depth, disciplined use of AI and an honest approach to solving unfamiliar problems.
We keep asking why and how until we reach the underlying behaviour. Candidates should understand what their framework, runtime, database, browser, cloud service or design tool is doing for them, how they work behind the scenes (and why), and how they would investigate a failure without relying on copying fixes from ChatGPT.
We expect candidates to use AI productively, but never as a substitute for judgement. They need to explain how they verify generated code, designs and tests, protect confidential data, catch unsafe assumptions and remain accountable for the result. We expect them to use MCPs, skills, and goals. So basically - we are looking for structured AI-assisted work, not vibe coding, vibe designing or vibe testing.
Strong candidates have a can-do mindset and will investigate, experiment and learn when the answer is not obvious. At the same time, they are fair and transparent about what they know, what they have not done before and when they need support.
Give us a few details about the role and your hiring plans. We'll assess your needs and get back to you with a practical next step.
We recruit machine learning engineers, AI developers, LLM engineers and AI research engineers across model architecture, training, fine-tuning, evaluation, retrieval and production inference.
Yes. We test data preparation, training objectives, optimisation, distributed training, full and parameter-efficient fine-tuning, evaluation, failure analysis and the trade-offs involved in adapting foundation models.
Our assessment can cover PyTorch, TensorFlow, JAX, Hugging Face Transformers, supervised fine-tuning, PEFT, LoRA, QLoRA, RLHF, DPO, distillation, RAG, model evaluation and MLOps.