Job Description

About the Team

The Platform Responsibility team is bringing disruptive innovation to content safety through the deep application of AI. We are building next-generation infrastructure that leverages advanced LLMs to reshape risk detection and governance. By engineering high-quality data pipelines and integrating AI into core decision-making, we drive exponential efficiency gains and precision. Working at the intersection of scalable engineering and technology, we empower global business growth while serving as the foundation of user trust.

About the Role

Under the spirit of TikTok Community Guidelines, the team leverages the capabilities of LLM and prompt engineering to deliver a large model Agent that can represent "absolute correctness", providing stable evaluation capabilities for TikTok content safety. This role will lead the team to refine strategy operations and prompt optimization to enhance the Agent's precise recall ability for risky content.

Responsibilities:

Prompt Engineering and Prompt Tuning

- Refined Prompting: Responsible for writing and debugging Prompts for specific risk scenarios. Maintain the Prompt version library to ensure the most appropriate instruction combinations are used for different scenarios.

- Few-shot Sample Library Management: Filter and maintain high-quality Few-shot samples (Examples). Ensure that the samples cover the latest underground industry variants and edge cases.

Badcase Attribution and Closed Loop Iteration

- Root Cause Analysis: Conduct in-depth analysis of cases that slipped through online (False Negative) or were wrongly removed (False Positive). Determine whether it is due to ambiguous Prompt instructions, outdated samples, or ambiguity in the Policy itself.

- In response to sudden public opinion events or new types of underground industry attacks, quickly adjust the Prompt strategy for interception, and convert the temporary strategy into long-term capabilities during the post-event review.

Knowledge Base and Evaluation Operations

- Knowledge Base Maintenance: For political and current affairs, and person-related risks, maintain the external Knowledge Base (Knowledge Base) of the RAG system to ensure that the Agent can identify the latest sensitive vocabulary.

- Daily Monitoring: Monitor and evaluate metrics daily, and output daily/weekly operation reports. When abnormal fluctuations in metrics are detected, issue timely warnings and troubleshoot the causes.