At Zuoyebang, I led three product iterations across local exam search, weekly learning reports, and a member-only knowledge graph. By redesigning content standardization and regional filters, I raised local exam search success from 32.0% to 78.6% and reduced search abandonment from 41.0% to 27.2% within 15 days.
I also created weekly progress summaries and a knowledge graph connecting textbooks, exercises, and videos. Membership conversion during final-exam review rose from 4.20% to 5.41%.
At iQIYI Lite, I used survey, interview, and behavioral data to identify friction in onboarding rewards and task redemption. I designed reward collection, a centralized task hub, and redemption paths, and owned instrumentation and experiment readouts; the work increased 7-day retention by 2%+, viewing time by 15%+, and task participation by 30%+.
On the MCN Commercial Contract Review Agent project, I designed an agent-assisted review workflow and built evaluation sets from anonymized contracts and scenarios. I tuned dual knowledge bases that reached 92.8% recall and 76.4% precision, while average legal review time during the pilot fell from 74.0 to 39.8 minutes.
