← All work
Responsible AI

Scaling content governance with policy and data

AI governancePolicyClassification
Improved AI classification accuracy by 10%.

The problem

As a content platform grows across regions, moderation decisions accumulate faster than informal policy can keep them consistent. The challenge is not only identifying harmful content; it is building a repeatable system that connects policy, human escalation, and model performance.

What I delivered

A cross-regional content-quality framework and global moderation policy for ByteDance / TikTok, along with an enterprise-scale AI content-governance program grounded in the analysis of more than 300 escalation cases.

What was hard

Policy must be specific enough to produce consistent decisions while remaining usable across languages, markets, and ambiguous edge cases. The escalation process also has to create learning data rather than merely resolve individual cases.

The result

The moderation policy became a company standard, and analysis of escalation patterns improved AI classification accuracy by 10%.

Running untrusted agents in production →Platform operations as agent-executable skills →Rules for the 80%, models for the tail →Right-sizing a fleet with a seasonal load curve →Shipping generative analytics before the playbook existed →Grounding a sales assistant in renewal knowledge →Building the operational view from zero →