Performance Calc

Problem

Pulling tables from many platforms, scoring and computing salaries by hand was mechanical, slow and error-prone — and every rule change rippled into payroll with no audit trail. I couldn't tolerate it, so I programmed it away.

Approach

1. Full pipeline: 8 platforms (WeChat Channels / Douyin / Kuaishou / Xiaohongshu / Bilibili / YouTube / Toutiao / Official Accounts) scraped 2–3× daily; auto-normalized, deduped across platforms (union-find merges the same video's variants), attributed to editor / writer, weighted-scored, ranked, and run through payroll (base tier + growth bonus + revenue share + penalty redistribution).
2. Rules as code, auditable: a single source of truth for scoring — hover any video or account to see the exact formula; any date range or account subset recomputes in seconds for weekly reviews.
3. AI performance assistant: a built-in Gemini function-calling chat that filters data, queries leaderboards and highlights records in plain language.
4. Integrated with AC (platformized): performance data sits next to the original scripts, with producer attribution synced across systems — retiring the Feishu tracking sheet entirely and dropping manual entry to zero.

Impact

• Compressed a ~10-hour monthly manual process to zero, saving managers ~150+ hours a year.
• Eliminated human scoring errors — accuracy and traceability were hard requirements since results feed payroll directly.

Tech stack