AI Research Engineer (Agentic Post-training) - 100% Remote Worldwide

Role overview Advance post-training methods that help AI models reason, plan, use tools, and complete multi-step tasks. You’ll research and build training and evaluation systems for tool-using agents, including models intended to run on resource-constrained devices and systems that work across text, images, and audio.

Responsibilities - Lead research and engineering work to improve post-training for agentic models and tool use. - Curate training examples such as tool-use trajectories and interactions with task environments. - Improve model behavior in areas including factuality, instruction following, reasoning, tool use, and coordination among agents. - Build and refine post-training pipelines, training workflows, evaluation suites, and benchmarks. - Diagnose weaknesses in tool-augmented models and develop methods to improve their performance. - Explore approaches to goal-directed behavior, self-correction, and reliable tool invocation.

Requirements - Deep knowledge of large language model architectures. - Substantial experience with post-training for agent workflows, including tool-use fine-tuning, function calling, or reinforcement learning from multi-turn feedback. - A hands-on, research-driven approach to developing and evaluating new training methods. - Ability to work across agent capabilities and, where relevant, multimodal systems and resource-constrained deployment. - Excellent English communication skills.

Nice to have - Open-source work on agent systems or tool use, including code, datasets, or models. - Research publications at leading AI conferences.

Benefits and work setup The position is described as fully remote and open worldwide.