Lead Engineer / Engineer, AI R&D (LLM Post Training), Q Team CoE

The Home Team Science and Technology Agency (HTX) is a statutory board under the Ministry of Home Affairs (MHA) which aims to pioneer innovation solutions and develop world class science and technology capabilities to transform and empower the Home Team in delivering safety and security for Singapore. The HTxAI Products team is committed to empower our Home Team Department (HTD)'s flagship projects through the seamless integration of AI technologies. From identifying opportunities to delivery, launch, and continuous enhancement, HTxAI aims to push boundaries and setting new standards in public safety. You will join the core team behind the Phoenix LLM – a family of Sovereign AI models trained right here in Singapore. We are an ambitious and dynamic team of Applied AI Scientists building state-of-the-art Large Language Models (LLMs) and Large Multimodal Models (LMMs) from the ground up. More about our work: http://go.gov.sg/phoenix-launch and https://arxiv.org/abs/2605.10391 In this role, you will have the unique opportunity to push the frontier of Sovereign AI while collaborating on joint Applied R&D projects with global AI labs. You will have access to large GPU clusters for distributed AI training. Responsibilities You post-train our LLMs for the agentic era: training it for our custom agentic workflows to call tools, carry out multi-step tasks. You will own the whole loop in a selected post-training domain: the reinforcement learning (RL) environments and harnesses the model learns in, the sandboxes where its actions run safely, the training tasks and the checkers that score them, and the fine-tuning and RL runs themselves. This is not a role building chatbots or applications on top of LLM APIs; your work forms the core agentic capabilities of our models. LLM Post Training: • Run SFT and multi-turn RL on our GPU cluster, tune training recipe, while mitigating catastrophic forgetting • Read agent transcripts to find where the model fails, and fix it through changes to environments, data, rewards or training recipes • Check with our product team that gains hold up in real use • Contribute to our technical reports and papers Environments and harnesses: • Build and scale tool-use and use-case-driven environments for agentic RL training, including tool servers and the automatic checkers that score the model's work • Maintain the harnesses that run our models through agentic tasks and benchmarks, used for both training and evaluation • Build containerised sandboxes on our GPU cluster, and own their security. That means code review, access controls - and testing that nothing escapes Requirements • Minimally Bachelor's degree in Computer Science, Data Science, Engineering, Mathematics or a related field, or equivalent experience • Hands-on experience fine-tuning or training LLMs (for example, SFT or RL) in PyTorch, through work, research, internships or your own projects • Strong Python, and comfortable working on Linux • Keen to learn and experiment in a fast-paced Applied AI R&D environment Good to have: • Experience with RL for LLMs (for example, GRPO, PPO or DPO) • Experience building agent harnesses, RL environments, evaluation harnesses, coding agents or sandboxed code execution • Distributed training experience on GPU clusters • Published papers or open source contributions We encourage you to apply even if you do not meet every point above. Applicants with 5 or more years of relevant experience will be considered for senior roles. **** All new hires are appointed on a two-year contract in the first instance and will be assessed and considered for permanent tenure over time, based on performance. As part of the shortlisting process for this role, you may be required to complete a medical declaration and/or undergo further assessment. All applicants will be updated on the status of their applications within 4 weeks upon closing of the advertisement.