Capabilities
SQL & BI as daily driver.
Hive, ClickHouse, and Aeolus (ByteDance's Tableau-equivalent). Run the numbers, build the leadership-review deck, no data partner in the loop.
Governance as platform, not policy.
Authored TikTok's first company-wide data access policy — and shipped it as a primitive inside the BI tool, not a PDF in a wiki.
Experimentation & pricing.
Stat-sig A/B design, geo-targeted pricing experiments, thousands of incentive campaigns automated in Python across 250+ Uber Eats markets.
Sales forecasting & unit economics.
Pipeline-driven forecasting methodology authored from scratch during TikTok's NA monetization launch. Modular pricing, radius optimization, real-time supply monitoring.
AI-agent layer for analytics.
Text-to-SQL, multi-step in-depth analytics agent, automated daily/weekly reporting agents — all in production for TikTok's global ad-revenue org.
Adoption & enablement.
The hardest part of analytics. Built the four-surface adoption engine — community, wiki, training, cookbook — that drove Analytics Center to majority adoption across thousands of commercial users.
Evidence
Analytics Center (TikTok).
Hybrid PM + PMM lead for ByteDance's internal AI-powered analytics platform for the TikTok For Business commercial org. Pivoted the platform strategy from compliance-focused data-access tool to AI-powered analytics partner — and the governance model from access control (who can see what) to definitional accuracy (everyone sees the same metrics). Drove majority adoption (~85%) across thousands of commercial users at sustained 4.0+ CSAT; ranked #1 among all internal sales tools. Shipped external-user access — moved ~1,000 partners, vendors, and agencies off sales-rep-mediated Excel exports onto self-serve dashboards with governance built in.
Text-to-SQL Production Agent.
Grounded on Analytics Center's verified-dataset layer. Authored a 100+ question ground-truth eval set, iterated prompts and retrieval grounding, lifted accuracy from ~30% to ~90%. Then stood up a self-improving flywheel — nightly batch testing, issue tagging, automated prompt refinement — that cut manual eval time by ~85% and drove a further ~70% post-launch error reduction.
In-Depth Analytics Agent.
Native to Analytics Center. Generates a reasoning plan, runs multi-step queries across selected datasets, self-recovers from query errors, outputs polished docs and web pages. Plus segmented daily/weekly reporting agents pushed to sales leadership each morning.
Global BI Platform Migration.
Led a six-month server migration of TikTok For Business's BI platform in response to U.S. regulatory requirements. Thousands of dashboards, hundreds of datasets, 100% migrated, zero cadence outage for the commercial org.
Dataset Consolidation.
Consolidated 100+ regional metric requirements across APAC, METAP, Europe, LATAM, and NA into ~20 core governed datasets. Replaced manual offline reporting with always-on, governed dashboards. Recognized with the 2022 ByteStyle Award (1 of 10 company-wide).
Uber Eats Pricing & Incentive Strategy.
Owned pricing and incentive strategy end-to-end across 250+ US/CAN markets on a multi-billion-dollar budget. Cut courier payout per trip by ~65%, lifted courier earnings by ~200%, lifted trips-per-hour by ~240%. Doubled NYC market share, grew weekly trips by ~95% — NYC #1 US/CAN, #3 global. Built the Start / Win / Stay courier lifecycle framework adopted as the US/CAN playbook.
Zentrading (solo).
Same diagnostic mindset, applied to retail equity research. AI agent layer doing multi-step reasoning over financial data, paid per analysis. zentrading.space
At scale
- Thousands of dashboards migrated, zero cadence outage
- Majority adoption on TikTok's internal AI analytics platform, #1-ranked of all internal sales tools
- ~30% → ~90% Text-to-SQL production accuracy via self-improving eval flywheel
- 100+ regional metric requirements → ~20 governed datasets
- Multi-billion-dollar pricing & incentive budget managed via SQL + Python automation
- ~85% reduction in manual eval time post-flywheel
Stack
SQL (Hive, ClickHouse) · Python · Aeolus (Tableau-equivalent) · Claude API · RAG retrieval · LLM orchestration · Lark / MagiBook · A/B testing frameworks · pandas · matplotlib