University of California, Berkeley

National Taiwan University

Feed AI team @LinkedIn.

We focus on make the LinkedIn feed more relevant and engaging.

LinkedIn

Engineering Manager, Machine Learning

Grew team from 2 engineers to 50+ engineers, dedicated to key ML problem spaces at TikTok US.

**Core Business Impact; All attributed via formal online A/B experiments and rigorous holdout**
Recommendation and Ranking:
+3% DAU, +6% watch time per user for TikTok main app
+5% DAU, +15% watch time per user for TikTok web app (tiktok.com)

E-Commerce Ranking:
+150% Gross Merchandise Value (GMV), bootstrapped E-Commerce business from zero to one

**Organization and Leadership**
Directly work with TikTok CTO and ByteDance General Managers on strategies for organization growth, building a localized MLE team in US.
Grew 2 engineer manager and hired another 2, each leading a team of 8~12 engineers.
Coached 15+ teammates to pass promotion and reach next levels.

**Trust and Compliance**
Built isolated data center dedicated to the US market, led large scale metric-parity investigation to reach 100% parity on North Star metrics (DAU, Retention, Watch Time Per User)
Contributed to key US safety problem spaces including election integrity, account safety, and direct message (DM) safety.

TikTok

Engineering Manager II, Machine Learning

Leading the Ads Entity Modeling and Targeting Foundations team. Expertise in Real Time Bidding, Ad Tech, Ad Funnel Optimization, and Applied Machine Learning

@Twitter Revenue Science, we combine machine learning knowledge with engineering practices to deliver the best advertising products for serving Twitter-scale traffic. My team focuses on modeling users, text phrases, images, and semantics for everything in the Twitter Ads world. These models could either power targeting products or be plugged into other parts of the ranking funnel as features.

Here are a couple of projects we work on/are responsible for:
- Integrating Relevance/Contextual Relevance in Ads Ranking
- Rebuilding the Twitter Candidate Generation/Early Ranking stack
- Auto targeting product for direct response advertisers
- Keyword lookalike; user lookalike
- Latent user interests

Frank Shyu