AI / ML Engineer

We're looking for an engineer who has trained and fine-tuned models and knows how to build and operate the infrastructure required to do it reliably. You'll own the model layer behind our assessment engine by fine-tuning models for risk classification and adversarial evaluation, optimising them for production, and building the training and inference infrastructure that keeps our assessment pipeline fast, scalable, and cost-efficient. You'll work directly with the founders.

Responsibilities:

• Fine-tune open-weight models for risk assessment, adversarial/jailbreak detection, and AI agent evaluation.
• Design and own the training infrastructure, including data pipelines (GT development), distributed training setup, experiment tracking, reproducibility, and operational workflows.
• Optimise models using quantisation, distillation, hyperparameter tuning, and inference optimisation while balancing latency, throughput, and cost.
• Stand up and manage GPU/compute infrastructure (cloud and/or self-hosted), including cost modelling and capacity planning.
• Design evaluation harnesses to measure model quality, robustness, and drift against the Ollive Risk Framework.
• Deploy, serve, and monitor models in production, including throttling, scaling, and failover handling.
• Collaborate with the Ollive AI Risk Labs and insurance team to translate risk methodologies into model behaviour.

Requirements:

• Strong Machine Learning and Deep Learning fundamentals.
• Hands-on experience fine-tuning LLMs/Foundation Models using LoRA, QLoRA, or full fine-tuning.
• Strong proficiency with PyTorch and the Hugging Face Transformers ecosystem.
• Experience with quantisation, hyperparameter tuning, and inference optimisation (including Speculative Decoding).
• Experience with distributed training frameworks such as DeepSpeed, FSDP, Accelerate, or similar.
• Experience setting up and operating training and serving infrastructure, including GPU provisioning, containerization, and model serving frameworks (SGLang, vLLM preferred).

Preferred Skills:

• Experience training models from scratch or working with 2B+ parameter models.
• Experience with large-scale cloud inference platforms such as AWS Bedrock and Amazon SageMaker, including quota and throttling management.

Good to Have:

• Familiarity with AI security or adversarial ML tools such as Promptfoo, Garak, Giskard, or similar.
• Exposure to regulated or compliance-heavy environments such as security, fintech, or insurance.