Machine Learning Engineer, LLM Post-Training

About the Role: We are looking for a hands‑on Machine Learning Engineer to drive the post‑training of our large language models, with a strong emphasis on reinforcement learning (RL). You will own the full post‑training stack — continuous pre‑training (CPT), supervised fine‑tuning (SFT), and RL — along with the data preparation that powers it. You will work directly with product and business teams to translate real‑world use cases into concrete training objectives and ship model improvements quickly.

Responsibilities: Lead post‑training of our LLMs across the full pipeline:continuous pre‑training, SFT, and reinforcement learning, with RL as the primary focus (e.g., RLHF, PPO, GRPO, DPO, and related methods).Design, build, and curate thedatathat drives each training stage — instruction/SFT datasets, preference pairs, reward signals, on‑policy rollouts, and rejection‑sampled completions — and define data‑preparation strategies tailored to specific business needs. Partner closely withbusiness and product stakeholdersto understand their scenarios, rapidly convert requirements into training plans, and deliver targeted model capabilities on tight timelines. Run large‑scale training onmid‑to‑large GPU clusters, applying distributed‑training techniques (data parallelism, FSDP, and where relevant tensor/pipeline parallelism) and tuning for throughput and stability. Build and maintainevaluation and reward/verifier pipelinesto measure model quality, prevent regressions, and ensure training‑serving consistency. Stay current with post‑training research and turn promising techniques into working, production‑ready code.

Requirements: Hands‑on LLM post‑training experience. You have personally run CPT, SFT, and RL training — withdemonstrated, practical RL experience(RLHF / PPO / GRPO / DPO or similar), beyond just launching training scripts. Strong data engineering for ML.You can independently design data‑preparation plans for a given business scenario — sourcing, cleaning, filtering, labeling strategy, and synthetic/preference data generation — to meet specific product requirements. Proven large‑scale GPU training ability. You have trained LLMs onmid‑to‑large GPU hardwareand are comfortable with distributed training and debugging at scale. StrongPyTorchfundamentals; working familiarity with frameworks such as Hugging Face TRL/Accelerate, DeepSpeed or FSDP, and inference engines like vLLM.Solid understanding of tokenization, attention, chat templates, and common failure modes in alignment/agent training. A bias towardfast iteration and business impact, with strong communication skills to work across research and product teams.