| layout | cv |
|---|---|
| title | Kevin Yen |
I build systems that connect millions of people with content they care about. Currently leading development of Yahoo's next-generation recommendation infrastructure—modernizing how we serve personalized content with LLM-powered systems that enable faster iteration and better user experiences.
How I work: I do whatever the project needs to succeed. Sometimes that means leading design discussions and coordinating across teams. Sometimes it means writing production code at 11pm to hit a launch deadline. Sometimes it means diving into research papers to figure out if a new technique will actually work at scale. I've led projects, written papers, debugged distributed systems, and refactored legacy codebases—whatever moves the work forward.
What I do: End-to-end applied ML and systems work, from initial exploration to production deployment. My work spans recommendation systems, ad tech optimization, content moderation, and infrastructure that enables other engineers to move faster. I'm comfortable being hands-on with code or stepping back to design architecture and coordinate execution.
Background: U.S. citizen with 8+ years at Yahoo, working across research and engineering. Deep expertise in ML, data science, and Python that naturally evolved into technical leadership across multiple efforts. MS from Ohio State, BS from National Central University (Taiwan).
Co-leading the design of Yahoo's next-generation recommendation system—rethinking how we serve personalized content at scale.
The opportunity: Build cloud-native, Python-first systems that integrate modern ML techniques including LLMs. Focus on faster experimentation cycles while maintaining performance standards and improving personalization quality.
What we're building: Complete stack from notebooks to data pipelines to production traffic handling. The system powers content feeds across multiple surfaces.
The Artifact integration: Following Yahoo's acquisition of Artifact (the AI news app from Instagram co-founders Kevin Systrom and Mike Krieger), I led technical aspects of integrating their technology into Yahoo's ecosystem. A 2.5-month sprint to adapt their backend to Yahoo's scale—the primary challenge was that Yahoo has significantly higher content volume and daily active users than Artifact was designed for. We refactored and connected services to internal systems while working directly with the founding team to understand their product and engineering approaches. Successfully launched as the new Yahoo News mobile experience.
Kevin Systrom and Mike Krieger previously founded Instagram; Mike Krieger is now Chief Product Officer at Anthropic
Traffic Optimization — Redesigned traffic handling for Yahoo's ad exchange. Developed ML-based throttling that replaces random rate limiting with value-aware traffic allocation, substantially improving efficiency while maintaining revenue.
Offline Estimation Framework — Built systems for running multiple experiments and estimating impacts before creating online test buckets. This enabled efficient communication with business teams and helped product owners fully understand potential outcomes before deployment.
Business Metrics Optimization — Designed and prototyped optimization models for key ad exchange metrics, providing clear impact analysis that informed strategic business decisions.
Ad Creative Policy Systems — Built classification models for political ads and policy violations that dramatically reduced manual review volume while enabling high-precision auto-clearing. Developed similarity-based creative hashing that further reduced review workload and improved training data quality through noisy label identification.
User-Generated Content Systems — Developed multiple generations of text and image classifiers for content moderation:
- Early BERT adoption (2019): trained from scratch with custom vocabulary—full pre-training on domain data, not fine-tuning a public checkpoint. Built ensemble systems that significantly outperformed external vendors.
- Transfer learning approaches for image classification
- Proactive toxic comment prediction to manage high-engagement content
- Search suggestion moderation with improved coverage of emerging problematic terms
Ad Creative Tools — Deployed optimization service enabling advertisers to improve creative quality, with latency improvements that made real-time recommendations viable in production.
Threat Detection Systems — Co-designed anomaly detection framework for large-scale network analysis to surface data-theft indicators—supporting security and compliance requirements.
Network Security — Designed data pipelines for enterprise-scale network events, enriching flow data for anomaly detection and forensics.
Authentication Security — Led development of deep learning models for Yahoo's membership security systems, reducing account-takeover risk across consumer accounts.
Systems & Tools — Contributions that improved team productivity:
- Established JupyterHub infrastructure with custom PySpark kernels for native notebook experience with remote Spark
- Integrated distributed training capabilities on internal Kubernetes clusters
- Led early cloud ML platform enablement that lowered barriers to experimentation
- Modernized ML monitoring dashboards and reporting
- Implemented CICD pipelines and conducted major codebase refactoring
- Built notebook automation for reproducible workflows
Currently using LLMs extensively in the recommendation system for feature generation, embeddings as training signals, and prompt engineering with systematic evaluation.
10+ peer-reviewed publications at venues including EMNLP, CIKM, IEEE TNNLS, and IEEE Big Data. Topics span text moderation, creative optimization, graph neural networks, and distributed systems.
4 patents and trade secrets covering content moderation systems, ad creative optimization, and security technologies.
See full publication list on Google Scholar.
Currently: Python, LLMs, GCP, AWS, Spark, TensorFlow, PyTorch, Kubernetes
Experience with: Scala, Java, TypeScript, JavaScript, C++, PHP, Rust, Go, Lua, SQL
Languages: English, Chinese
- 2023–present — Senior Principal Research Engineer, Yahoo (NYC)
- 2020–2023 — Principal Research Engineer, Yahoo (NYC)
- 2019–2020 — Senior Research Engineer, Yahoo (NYC)
- 2018–2019 — Research Engineer, Yahoo (NYC)
- 2017–2018 — Associate Research Engineer, Yahoo (NYC)
- 2016–2017 — Part-time Contractor, Yahoo (Remote)
- Summer 2016 — Research Intern, Yahoo (NYC)
- 2015–2017 — MS Computer Science, Ohio State University
- 2010–2014 — BS Computer Science, National Central University (Taiwan)