Lishu Liu

A. Technical Leadership & Full-Stack ML Ownership

• Architected and delivered a next-generation ranking model that unified GBDT and neural network (TTSN) approaches, strategically selecting each for specific sub-tasks based on latency and performance trade-offs.
  Result: Achieved a +1.5% lift in user engagement and a +0.8% increase in session time across a user base of hundreds of millions, the largest quarterly gain for the team.
  Pioneered the use of a Value Model to optimize for long-term user value over short-term engagement, defining the new optimization target and aligning cross-functional partners on the strategy.
• Led a major infrastructure integration by adopting a state-of-the-art serving stack from Instagram, adapting it for Feed's scale and latency requirements.
  Result: Reduced p99 model serving latency by 40ms, enabling the deployment of more complex models and contributing to a +0.5% reduction in user drop-off.
• Drove a cross-team initiative (Wikiathon) to overhaul feature freshness, implementing intelligent impression capping, and dynamic diversity mechanisms.
  Result: Reduced content over-exposure by 22% and improved user-reported satisfaction metrics by +5%, directly addressing a key product integrity goal.

B. Strategic Impact & Modeling Innovation

• Solved the "Cold Start" problem for new posts and creators by developing a novel ranking framework using collaborative filtering and lightweight retargeting signals, deployed as a "Starter Pack" in the ranking stack.
  Result: Improved the visibility of new content by 35% and increased the retention rate of new creators by 15% within their first week.
• Owned the end-to-end feature engineering lifecycle, from conceptualization to production. Introduced contextual engagement signals (e.g., time of day, user session context) that became foundational to the "Quick Access" and enriched ranking models.
  Result: These features contributed to a +2% increase in relevance scores (as measured by human evaluation) and were adopted by 3 other teams.
• Championed and implemented a platform-wide privacy compliance project, analyzing and modifying feature pipelines and model architectures to adhere to new age restriction and data privacy rules.
  Result: Successfully de-risked a high-priority legal requirement, ensuring compliance for 100% of affected users without a statistically significant regression in core engagement metrics.

C. Platform Influence & Mentorship

• Established a new performance monitoring paradigm by designing and building a real-time dashboard for ranking health metrics (latency, freshness, diversity), reducing the mean-time-to-detect (MTTD) regression issues from hours to minutes.
  Result: This dashboard is now the standard for the >50 engineer ranking org, used daily for on-call and product launches.
• Acted as a technical mentor and onboarding lead for the team; directly onboarded and mentored 4 new E4/E5 engineers and 1 intern, with two mentees receiving top performance ratings.
  Result: Scaled team effectiveness and reduced their ramp-up time to full productivity by an estimated 30%.
• Recognized as a domain expert; frequently invited to give tech talks on ranking architecture and model evolution to audiences across the company, influencing the technical roadmap of adjacent teams.

• Drove User Engagement with a Novel Recommendation System
  Owned the end-to-end development of a Collaborative Filtering model from scratch to personalize the mobile app home page.
  Result: Achieved a 29% increase in user engagement, demonstrating a strong ability to translate ML theory into product impact.
• Pioneered Infrastructure Efficiency and Cost Savings
  Led a proof-of-concept (POC) with Databricks that successfully improved team productivity and reduced project cloud costs by 50%.
  Optimized large-scale Spark applications and ETL pipelines, reducing total runtime by 30-33% and improving team iteration speed.
  Proactively shared expertise in Spark and distributed systems beyond immediate team responsibilities, acting as a resource for others.
• Mastered New Technologies to Solve Complex Problems
  Designed and implemented a clustering algorithm enhanced with business rules to directly improve the performance of the Store Page Ranking model.
  Learned Scala in one week to write and deploy production-grade Spark jobs, showcasing an ability to rapidly acquire new skills to meet project demands.
• Built End-to-End ML Solutions for Ambiguous Problems
  Developed a highly accurate Conditional Random Field (CRF) model to automatically extract structured product information (title, price, image) from unstructured HTML for a Price Comparison Tool.
  Took initiative to build a Chrome extension to streamline and accelerate the data labeling process, unblocking the team and improving data quality