Senior Member of Technical Staff (Post-Training)

About Company

Founded in 2022, XG Tech is driving the future of smart vehicles. Its mission is to empower the digital transformation of automobiles, moving from distributed computing to a centralized, cross-domain platform.

XG Tech focuses on the intelligent cockpit—the next frontier of differentiation—while seamlessly integrating advanced driving systems. By reimagining cars as mobile living spaces, XG Tech aligns with the evolving trend of vehicles becoming the “third living space.”

Role Summary

As a Post-Training Research Engineer, you will research and develop post-training techniques to enhance the capabilities of on-device language models, leveraging methods such as Supervised Fine-Tuning (SFT), Reinforcement Learning (RL), and Online Policy Distillation (OPD). You will optimize model performance, reasoning capabilities, and task execution to bring lightweight on-device models closer to frontier-level performance while maintaining efficient inference, low latency, and reliability on resource-constrained devices.

Key Responsibilities

• Own the post-training pipeline end to end: data construction, teacher selection, loss and objective design, and the unglamorous work of making gains survive quantization and deployment.
• Move frontier capability into small models for on-device use: SFT for the cold start, RL for the behaviours that imitation cannot teach, and OPD when a stronger teacher can still be squeezed. Know which lever to pull, and when one of them stops paying.
• Build the data and generation pipeline that produces post-training corpora at scale, with quality control that holds up under scrutiny rather than a pass/fail eyeball.
• Push beyond imitation: study whether reasoning, tool use, and long-horizon behaviour transfer through post-training, rather than only surface style and formatting.
• Design the evaluation loop that tells you whether a run actually improved the model, including regression checks.
• Work with the routing, evaluation, and embedded teams on what "good enough on device" means in practice — the constraint is real, and it is the point of the role.
• Publish when results are defensible; keep a clear internal record when they are not, and say so.

How you will stand out

• Strong academic background in Machine Learning, NLP, Computer Science, or a related field, with a PhD or equivalent research experience preferred.
• Deep understanding of LLM post-training techniques, including Supervised Fine-Tuning (SFT), Reinforcement Learning (RL), and Online Policy Distillation (OPD).
• Hands-on experience in model training and optimization, particularly in improving reasoning capabilities, task performance, and knowledge transfer in smaller language models.
• Strong engineering skills in building and managing training pipelines, distributed training infrastructure, experiment tracking, and systematic model evaluation.
• Demonstrated research or engineering achievements through publications, technical competitions, open-source contributions, training frameworks, or successfully developed and deployed AI models.
• Ability to independently design experiments, analyze training outcomes, and systematically improve model performance through data-driven approaches.

Nice to have:

• Experience optimizing models for on-device deployment, including quantization, KV-cache optimization, and memory-efficient inference.
• Experience with advanced post-training techniques such as OPD, Reinforcement Learning with Verifiable Rewards (RLVR), and preference data construction.
• Familiarity with agentic post-training, including tool-use learning, multi-turn interactions, and agent trajectory optimization.
• Experience developing compact language models with strong reasoning, instruction-following, and task execution capabilities under resource constraints.

XG Tech is committed to providing equal employment opportunities by country, state, and local laws. XG Tech does not discriminate against employees or applicants based on conditions such as race, color, gender identity and/or expression, sexual orientation, marital and/or parental status, religion, political opinion, nationality, ethnic background or social origin, social status, disability, age, indigenous status, and union.