Applied Scientist

Most companies are building AI products on top of foundation models. We’re looking for scientists who can make those models better. We’re an exceptionally well-funded applied AI startup deploying frontier AI into governments, healthcare and critical industries. Rather than training foundation models from scratch, our work focuses on improving model capability through post-training, optimisation and applied machine learning—building systems that outperform off-the-shelf models on real customer problems. This is a deeply technical modelling role for scientists who enjoy experimenting, measuring and improving model performance. The Role You’ll own the modelling behind production AI systems. That means designing experiments, improving model behaviour, building evaluation frameworks and developing novel approaches to increase capability—not simply integrating existing APIs. You’ll work across areas including: Post-training and model adaptation Reinforcement learning and hill-climbing optimisation Preference optimisation and reward modelling Prompt optimisation and automated search Model evaluation and benchmarking Inference-time optimisation Agentic reasoning and planning Multimodal modelling Your work will directly determine how our AI systems perform in production. What You’ll Be Doing Develop novel modelling approaches that improve frontier model performance on complex real-world tasks Build post-training pipelines that increase capability, robustness and reliability Design hill-climbing and iterative optimisation algorithms to continuously improve outputs Create evaluation datasets, reward functions and automated benchmarking systems Run large-scale experiments to understand model behaviour and identify performance gains Partner with engineering teams to deploy improvements into production Translate cutting‑edge ML research into measurable customer impact We’re Looking For We’re specifically looking for Applied Scientists who have owned the modelling , rather than engineers who have primarily built applications around existing models. You’ll likely have experience with several of the following: Designing or improving ML models rather than simply consuming them Post-training techniques including SFT, RLHF, DPO, GRPO or related optimisation methods Reinforcement learning, hill-climbing or iterative search algorithms Building evaluation frameworks and reward models LLM adaptation, optimisation or fine-tuning Strong Python and modern ML frameworks (PyTorch, JAX or TensorFlow) Running rigorous experiments and using data to drive model improvements We’re particularly interested in people who enjoy asking: *\"How can we make this model perform materially better?\"* rather than: *\"How can we build an application around this model?\"* Why Join? Work on some of the hardest applied AI modelling problems in industry Shape how frontier models are adapted for real-world deployment Join a small team of exceptional scientists and engineers with significant technical ownership Work with the latest frontier models, datasets and infrastructure $250,000–300,000 base salary plus meaningful equity Hybrid working in San Francisco (3 days per week) If you’ve built the models—not just the products around them—and enjoy pushing frontier AI beyond its default capabilities, we’d love to hear from you. #J-18808-Ljbffr