You will work on world models, diffusion models, and multimodal AI, developing novel architectures, training paradigms, and large-scale distributed training/inference infrastructure. The lab creates next-generation foundation models to understand, predict, and interact with the physical world, targeting applications in robotics, autonomous systems, and other domains.

Work with world models, diffusion models, transformers, video generation, multimodal learning, and large-scale ML systems. As part of Odyssey's applied research team, you'll advance causal, multimodal world models for robotics, science, and other fields, building from hypothesis through working prototypes and shipping results into production models.

You will work with generative AI, autonomous agents, and real-time rebalancing systems across TikTok, Meta, and Snap to optimize paid acquisition. This role builds a first-of-its-kind marketing engine for EQUALS, a London-based music social network with 1 million monthly active users.

DeepMind★5.0 (2)· London

Work with AI model training, testing, evaluation, and tuning processes, and LLM release cycles and timelines. The role is on the Gemini Release Evaluations team, focusing on creating new SOTA datasets for frontier models and improving understanding of static evaluations versus live production traffic.

Pacific Life★3.7 (310)· London

Python, R, SQL, machine learning, and cloud platforms such as AWS or Azure. The Strategic Analytics team researches mortality, morbidity, and disability risk using large longitudinal datasets to inform underwriting, risk assessment, and product development.

Work with AI models, large language models, evaluation frameworks, and datasets to design and implement release evaluations. You will identify gaps in existing evaluations, create new state-of-the-art datasets, and improve understanding of the relationship between static evaluations and live production traffic.

Research on multi-agent reinforcement learning for autonomous underwater vehicles using deep learning, Python, and PyTorch. The role advances AI for Arctic security, enabling cooperative autonomy under severe communication and localisation constraints.

Working with Python and modern machine learning frameworks, you'll research Agent systems, Large Language Models, and reinforcement learning — focusing on agent memory, long-term reasoning, self-evolution, and agentic RL. The team builds sustainable, generalisable, evolvable autonomous AI agents for complex environments, advancing research toward AGI.

You will work with scripting languages, pipelining tools, and machine learning, focusing on adversarial robustness, data influence, watermarking, and GenAI audio/video models. The team accelerates research on trustworthy, robust ML, including the provenance of synthetic media and data trustworthiness, with SynthID as a public example.

You will work with Python and ML frameworks like PyTorch to conduct original research in interpretability. The team advances the science of how large AI systems work, developing techniques to understand, debug and steer model internals.

You will design and train custom deep-learning models and fine-tune large language models using Python, TensorFlow, or PyTorch. The work involves leading research for Kangal AI, a venture building frontier AI for governments and security clients. You will own the entire model stack and scale a research team as a founding hire.

You will work with Python, TensorFlow or PyTorch, and large language models, building custom deep-learning models and training/fine-tuning LLMs on a proprietary dataset. Kangal AI builds frontier AI for governments, national security, and private corporations. As a founding hire, you will own the model stack and scale the research into a team you lead.

You will work with machine learning and object-oriented programming languages to extract signals from vast market-data datasets, supporting alpha generation and cross-asset trading teams. The team operates within a leading market maker and liquidity provider, using ML to improve daily performance and build tools for traders and engineers.

Founding AI Research Scientist role at Blue Wolf Digital: design, train, and fine-tune custom deep-learning models and proprietary LLMs using PyTorch, owning pipelines end-to-end. The company builds frontier AI for governments, national security, and private organizations, and as the founding hire you'll set research strategy with the CEO, then hire and lead the research team.

Canva★4.3 (32)· London

You'll work with Python, PyTorch, vision-language models (Gemma, Qwen-VL, SmolVLM, MiniCPM-V), optimization techniques (quantization, pruning, distillation, hardware-specific compilation, task-specific fine-tuning), and on-device runtimes (Core ML, LiteRT/TFLite, ONNX Runtime, ExecuTorch). The Video Storytelling team builds models and systems behind Canva’s video AI, partnering with the Edge AI group to explore on-device deployment. This internship optimizes a video-capable VLM for on-device intelligent captioning on consumer phones.

Develop predictive deep learning models using multimodal wearable data (heart rate, sleep, activity) and voice biomarkers. You will work on the EMBRACE research programme, a global study aiming to advance precision maternal and early childhood health through AI and personalised interventions.

Working on robot learning, VLA, and world models using modern deep learning and ML. This founding role at a UK Physical AI startup researches models enabling robots to understand, learn, and act in the real world, with experiments on real hardware. You’ll lead research direction as part of a small, highly technical founding team.

Canva★4.3 (32)· London

Work on reinforcement learning and diffusion modelling for Canva’s AI team, using policy-gradient RL (GRPO, PPO, DPO), VLMs, LoRA, reward modelling, and distillation. You’ll design VLM judges for RGBA layer decomposition, build evaluation methods, and turn them into reward functions, with PyTorch and multi-GPU training (FSDP, DeepSpeed) on a live, industry-scale project.

DeepMind★5.0 (2)· London

Use scripting languages, pipelining tools, and ML techniques (e.g., adversarial robustness, data influence, watermarking) to work on strategic projects ensuring trustworthy, robust, and reliable machine learning—such as the SynthID system for synthetic media provenance.

You will work with Python and ML frameworks such as PyTorch to conduct research on interpretability, developing techniques to understand, debug, and steer large AI models. Goodfire is a research company advancing the science of how AI systems work, and you will collaborate with a small, mission-driven team of scientists and engineers in London.

Use scripting languages, pipelining tools, and ML techniques to advance research on robustness, provenance of synthetic media, and data trustworthiness. The team drives strategic projects like SynthID to enable trustworthy, reliable machine learning.

Canva★4.3 (32)· London

You'll use VLMs, diffusion/flow-matching, policy-gradient RL (GRPO, PPO, DPO), and PyTorch with multi-GPU training (FSDP, DeepSpeed) to build VLM-based evaluators and reward models for RGBA layer decomposition and layered generation. The project feeds into Canva's AI-powered design platform.

You will apply machine learning techniques to problems of robust, trustworthy model deployments, working with scripting languages, pipelining tools, and GenAI audio/video models. You will collaborate with research scientists and engineers on projects like SynthID, focusing on provenance, adversarial robustness, data influence, and watermarking. This is an individual-contributor research scientist role in a mission-driven team.

Faculty★0.0 (16)· London

Research Scientists execute technical research in AI safety, using Python with PyTorch to study and steer transformer-based LLMs via white-box and black-box interpretability techniques like mechanistic interpretability and steering vectors. They collaborate with senior scientists at Faculty, a company that builds and deploys human-centric AI for over 350 global clients.

Sonar★5.0 (2)· London

You will work with Python, core ML frameworks, and post-training techniques including GRPO, PPO, RLVR, SFT, PEFT, and preference/safety alignment (Rust or SonarQube flagship languages are a plus). You will join a cross-disciplinary team developing advanced products that post-train models to power enterprise agentic coding practices, ensuring code meets enterprise standards.