Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Drive core technology development for large language model code direction, continuously optimizing code comprehension, reasoning, and generation capabilities.
Focus on improving code comprehension, reasoning, and generation capabilities in real-world production codebases, improving TikTok service code performance and privacy compliance.
Explore Code Agent capabilities suitable for actual business production environments, and improve TikTok R&D efficiency....
Drive core technology development for large language model code direction, continuously optimizing code comprehension, reasoning, and generation capabilities.
Focus on improving code comprehension, reasoning, and generation capabilities in real-world production codebases, improving TikTok service code performance and privacy compliance.
Explore Code Agent capabilities suitable for actual business production environments, and improve TikTok R&D efficiency....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals....
Participate in defining and continuously improving data quality standards, judgment rules, and acceptance criteria for large model training and evaluation scenarios.
Review data outputs against quality standards, identify issues, drive correction loops, and ensure accuracy and consistency.
Assess complex, ambiguous, and boundary cases, and turn decisions into reusable judgment rules and knowledge assets....
Prompt Engineering: Design and deploy advanced prompting strategies, including Chain-of-Thought (CoT), Meta-Prompting, and Least-to-Most Prompting, to precisely guide model behavior.
AI Tool Integration: Maintain and leverage automated Root Cause Analysis (Auto-RCA) tools and automated Prompt Agents to quickly identify the causes of model failures and generate potential solutions within minutes.
Logic Diagnosis: Conduct deep-dive analysis into model hallucinations and classification errors. You will identify whether failures are caused by tokenization issues, logical gaps, or variance driven by temperature settings....
Under the guidance of a mentor, participate deeply in the development of core TikTok Search scenarios such as the search results page. Work closely with Product, Design, and Algorithm teams, and drive feature iterations through A/B testing and data analysis to continuously optimize the search experience for hundreds of millions of users.
Based on multi-platform technology stacks such as Lynx, Android, iOS, and Server, participate in the development and optimization of search business modules, and systematically build cross-platform engineering experience through real-world challenges such as performance bottlenecks and stability assurance.
Integrate AI Agents into daily R&D workflows—from coding assistance and intelligent debugging to automated testing. Participate in the exploration and practical application of AI tools in search business scenarios, and be encouraged to bring forward your own innovative ideas....