Resume
Experience
Jan 2026 to Aug 2026
Data Scientist, Agentic Engineering · GrowthGPT
Remote
- Built the analytics framework for a B2B LLM ad agent that diagnoses cross-channel campaign performance; raised diagnostic accuracy from 65% to 90%, reduced token use, and cut latency by 2–3 min with the base model unchanged.
- Defined unified cross-channel field and metric specifications and a semantic layer that maps ambiguous user requests to structured query scopes, replacing high-latency direct API calls and unconstrained text-to-SQL over raw tables.
- Built a regression suite from production failure cases in conversation logs; codified analytical SOPs into reusable Agent Skills.
Nov 2022 to Mar 2025
Data Scientist, User Growth · miHoYo (HoYoverse)
Singapore
- Optimized Genshin Impact user acquisition by profiling devices via chipset performance; excluding low-LTV tiers in Google Ads drove a 5% ROI lift and purchase rates in three weeks, leading to the scaling of postback suppression across all major ad networks.
- Measured the incremental lift of regional advertising with synthetic control where randomized tests were not feasible; results guided quarterly allocation of the game’s ~$100M annual ad budget across media networks.
- Restructured SKAdNetwork (SKAN) conversion values around onboarding and progression milestones, lifting attributed ROAS 13%.
- Developed a creative evaluation framework leveraging CPC and downstream conversion metrics; implemented data-driven creative revisions that lifted install conversions by 8%.
Nov 2021 to Nov 2022
Data Scientist, Tencent QQ · Tencent
Shenzhen, China
- Replaced a rule-based feed recommender with an incrementally trained LightGBM model utilizing weekly user-event rollups and real-time behavioral features; drove a 2-percentage-point lift in recommendation-card CTR validated via A/B testing.
- Ran survival analysis on existing users to uncover behavioral patterns and build user segments; reactivation campaigns for high churn-risk segments significantly raised active users versus control in an A/B test and added ~4 minutes of daily app time per user.
- Spearheaded cross-functional data governance initiatives alongside engineering and product teams; established standard app event-tracking specifications that significantly enhanced data reliability across the department.
Mar 2019 to Sep 2021
Data Scientist, Claim Home Office · GEICO
Washington, D.C.
- Built a two-stage hurricane loss model (logistic regression for vehicle-level claim probability, XGBoost for 30-day policy-level loss) on historical hurricane tracks, insured-vehicle geospatial exposure, and storm characteristics; automated it with National Hurricane Center forecasts, serving results via an internal API and department alerts for claims staffing.
- Built a Hadoop entity-resolution pipeline (deterministic, fuzzy, and probabilistic linkage) that unified millions of nationwide U.S. claims records with no shared join key.
- Guided customer service strategy improvements by identifying critical NPS drivers through random forest feature importance analysis.
Education
M.S. Information Management (Data Science) · University of Washington
M.S. Civil and Environmental Engineering (Transportation Risk Analysis) · Rutgers University
Skills
- Experimentation & Causal Inference
- A/B testing, difference-in-differences, synthetic control, metric design
- Growth & Ads Measurement
- marketing mix modeling (MMM), attribution, SKAN, LTV/pLTV modeling, CAC/ROAS analysis
- Machine Learning & Statistics
- XGBoost, LightGBM, logistic regression, random forest, survival analysis, recommender systems
- Tools & AI
- Python, SQL (Hive, BigQuery), Spark, Tableau, LLM agents, MCP, Agent Skills, LLM evaluation, Claude Code, Codex