Aria Li

AI Researcher

Fremont, California, United States5 yrs 10 mos experience

Most Likely To SwitchHighly Stable

Key Highlights

Architected recommendation engines for 100M+ gamers.
Achieved $8.9M revenue impact through ML models at PayPal.
Developed first chatbot in PayPal with high BERT score.

Stackforce AI infers this person is a Machine Learning Engineer with expertise in Gaming and Fintech industries.

Contact

Skills

Core Skills

Machine LearningData AnalysisRecommendation SystemsNlp

Other Skills

PythonMySQLreal-time re-rankerembedding optimizationfeature serving architecturemulti-task modelcontent metadataRAGchatbot developmentdata scrapingJavaScriptWordPressnatural language process

About

I design, build, and operate machine learning systems to serve 100M+ gamers at scale. As a Senior Machine Learning Engineer at Sony PlayStation, I architect the recommendation engines that power personalized game discovery. My systems analyze billions of player interactions to surface the right game at the right moment—whether it's a blockbuster release or a hidden indie gem. Here's what I'm building: • PlayStation's first real-time re-ranker with hybrid voter architecture (relevance + dynamic recency scoring) • I also explored LLM-augmented recommendations that blend behavioral signals with natural language understanding • An embedding-as-a-service platform adopted by other models across the PlayStation ranking systems • I bridged the gap between research and production by: Building pragmatically: Saved $761K through two-tower embedding optimizations Beyond gaming, I bring unique FinTech ML experience from PayPal: • $8.9M revenue impact from merchant LTV/churn models • LLM & NLP: Led end-to-end development of a RAG-based Slack chatbot --- first chatbot in PayPal, fine-tuning Llama-2 and achieving 0.83 BERT score for financial workflow. Impact: Reduced merchant onboarding support queries by 30%, saving ~200 engineering hours/month

Experience

5 yrs 10 mos

Total Experience

1 yr 9 mos

Average Tenure

2 yrs 2 mos

Current Experience

Netflix

Machine Learning Engineer

Sep 2025 – Present · 9 mos

Integrated foundation‑model–based entity signals into MESA (title–profile affinity scores and member profile embeddings) so the system can understand which specific messages each member is most likely to engage with.
This is a key step in shifting from message‑type–based targeting to deep, entity‑level understanding of message content, which improves Meaningful Message Action Rate by ~0.5%. MyList Add, Primary Title Play, and Remind Me Add all showed significant lifts.

machine learningPythondata analysisMySQL

Playstation

Senior Machine Learning Engineer

Apr 2024 – Present · 2 yrs 2 mos · San Francisco, California, United States · Hybrid

1. PlayStation's First Real-Time Re-Ranker System
Designed and built the platform's first production re-rank system with intelligent content freshness awareness:
Relevance Voter
Multi-task model combining gameplay patterns, store interactions, and content metadata
Unified understanding of players and games across PlayStation ecosystem
Recency Voter
Content-aware time decay framework
Custom rules for games, DLCs, and bundles
Strategically aligned with PlayStation's release cycles
Key Innovations
First real-time re-ranker system for PlayStation
Novel content-type-aware freshness scoring
Foundation for future recommendation improvements
2. Performance-Optimized Recommendation System
Redesigned two-tower serving architecture achieving:
36.7% latency reduction (30ms → 19ms p99) through online dot-product simplification
$761K annual cost savings from
Model API optimization: Reduced Seldon calls from 50k to 2.5k QPS (95%↓)
Optimized Feature Serving Architecture
Redesigned feature pipeline from real-time feature computation to pre-computed embedding
Technical Impact:
▸ Pioneered embedding sharing through internal feature store, adopted by downstream models
▸ Served as blueprint for 2 other ranking systems

machine learningreal-time re-rankerembedding optimizationfeature serving architecturerecommendation systems

Paypal

2 roles

Sr, machine learning engineer at PayPal

Promoted

Sep 2022 – Apr 2024 · 1 yr 7 mos

1. Merchant Product Recommendation--- Developed a personalized recommender system for merchant activation using merchant profiles, transaction behavior, product usage, descriptions, and categories.
2. Slackbot --- Led the development of a cutting-edge chatbot integrated with PayPal products within the Slack platform. Experimented with state-of-the-art language models, including fine-tuning the Llama2 model and employing the Retrieval-Augmented Generation (RAG) approach. Successfully achieved a BERT score of 0.83 using the RAG approach, demonstrating significant improvement in conversational quality and relevance.
3. Churn Detection Model, providing churn probability and actionable reasons. Solution adopted by strategy team for retention program, reduced small and medium merchants annual churn rates.
4. Customer Value Prediction Model, leveraging merchant profile and behavior as features to detect high-value merchants for targeted promotions and incentives.
5. Email Engagement Model, Leveraged Uplift model to predict the effect of sending email offers to customers.
6. Developed a cutting-edge Privacy Embedding Framework (Research Project), designed to encrypt sensitive customer information while preserving the performance of modeling and analytics.The framework demonstrates innovation and potential, evidenced by its submission as PayPal patent.

machine learningNLPRAGchatbot development

Machine Learning Engineer

Dec 2020 – Aug 2022 · 1 yr 8 mos

Ansell

Machine Learning Scientist intern

Jul 2019 – Dec 2019 · 5 mos · Greater New York City Area

Built a prototype to improve sales by customizing product recommendations for Ansell’s customers based on product feature dataset and user historical activities
Conducted feature embedding on product description by TF-IDF transformation. Reduced dimensions of long feature list via Auto-Encoder and reduced storage compacity by 90%
Built two Recommender system for warm users: item-based system using weighted similarity matrix, Collaborative-Filtering based system, constructed popularity-based system for cold users
Assisted A/B testing design and implementation for customizing recommendation to the client. Developed a data scraping pipeline through Mysql database, Turbodbc and Google Analystics and generated the performance report and presented to data science team

machine learningrecommendation systemsdata scraping