About the company
About OKX OKX is a leading crypto trading app, and a Web3 ecosystem. Trusted by more than 20 million global customers in over 180 international markets, OKX is known for being the fastest and most reliable crypto trading app of choice for investors and professional traders globally. Our Singapore office is a Product and Engineering hub and we are in the progress of expanding our teams in Singapore for the continuous growth of our global business. We build and maintain core trading platform with millions of daily active users. Design, Product and Engineering teams work cross-functionally to identify customer needs, and ship high-quality new features through fast iterations.
Job Summary
What You’ll Be Doing
📍Lead and execute the full post-training pipeline for large language models (LLMs), including supervised fine-tuning, preference optimization, and reinforcement learning–based methods. 📍Design and implement advanced training paradigms such as DPO (Direct Preference Optimization) and GRPO (Generalized Reward Policy Optimization). 📍Develop domain-specific data recipes, curation strategies, and augmentation pipelines to optimize task performance. 📍Conduct post-training of specialized small models from scratch, including architecture selection, dataset construction, and optimization strategy. 📍Build and refine Reward Models to support alignment and downstream optimization. 📍Design and implement RLAIF (Reinforcement Learning from AI Feedback) closed-loop systems. 📍Optimize inference efficiency and deploy models using low-latency serving frameworks such as vLLM and SGLang. 📍Evaluate model performance using both automated benchmarks and human/AI feedback loops. 📍Collaborate with research and infrastructure teams to productionize training and deployment workflows.
What We Look For In You
📍Bachelor's in Computer Science, AI, Machine Learning, or related fields with at least 8 years of industry experience. 📍Strong hands-on experience across the full post-training pipeline for large models. 📍Deep familiarity with preference learning and alignment techniques, including DPO, GRPO, and RL-based post-training methodologies. 📍Proven experience designing domain-specific data strategies and training methodologies. 📍Experience training and post-training specialized small models from scratch. 📍Solid understanding of reinforcement learning fundamentals and their application to model alignment. 📍Experience deploying models in low-latency production environments using frameworks such as vLLM, SGLang, or similar.
If you’re passionate about blockchain and decentralized technologies, explore more opportunities in web3 and cryptocurrency careers.
