Jun Yan Email & Phone Number

California, United States

San Francisco, California, Us

2022 Oct - now: Uber Ads2022 April- Sept: UberAI, uMetric

San Francisco, Ca, Us

Shop Recommendation, Personalization and Ranking

San Francisco, Ca, Us

I lead a central data science team at Demandbase. Our team is responsible for all DS initiatives and projects in Demandbase. We work cross functionally with UI/UX designers, product managers, project managers, data engineers, application developers, devops and custom success managers to deliver the best account based experience using data, ML and AI. We are currently working in three areas: - Build new predictive analytics features for the DB1 platform. The Qualification and Pipeline Predict scores were build with machine learning algorithms and deployed to hundreds of our customers. - Improve traffic identification of mapping hundreds of billions of IPs and cookies to companies. - Improve B2B advertising performance and deliver new B2B advertising products.

San Francisco, Ca, Us

• Lead the Data Science effort in B2B advertising team. My team works cross functional with product/design, data engineering and application and built inventory forecasting, audience segments, click through rate (CTR) model, engagement model, pacing and pricing strategy, experimentation and AB test, intent model; as well as designed campaign KPIs and built performance dashboards. • Built a scalable v2 version of CTR model and resulted in 70% increase in CTR and percentage of poor performing campaigns dropped from 25% (v1 model) to 3% (v2 model). • Introduced deep learning models for real time contextual targeting. Resulted in substantial improvements in targeted page relevance to our customers.

San Francisco, Ca, Us

• Built collaborative filtering algorithm to identify and target high intent buyers from the 150 billion B2B intent signals every month, and resulted in 30X ROI comparing to other ads agency.• Implemented click through rate (CTR) model for real time bidding (RTB) in Golang. • Developed and implemented Pricing strategies for targeting.• Built data visualization and dashboards for KPIs for monitoring purposes and for consuming by other stakeholders. • Built monitoring and alert system for the ML products I built. • Tech stack used : Apache Airflow for data pipeline and job scheduling; Google Bigquery for data ETL; Jyputer Notebook, Scikit-learn, Pandas, Matplotlib for model prototyping with small sampled data; Spark MLLib, PySpark, Qubole and Amazon S3 for model prototyping for large data, Scala and Google DataProc for productize the machine learning algorithms, Google Datastudio for building and monitoring KPIs, Grafana for monitoring and alerting, Kibana and NewRelic for logs and trouble shooting.

San Francisco, Ca, Us

• Developed algorithms to rank keywords by their relevance to customer products and implemented the solution as Keyword Relevance API. This API powers all Demandbase AI Solutions including ABM Platform, Targeting, Conversion and Real Time Intent.• Increased relevance from 68% to 98% for displayed webpages using natural language processing techniques matching billions of webpages with keywords.• Built random forest classifier on Spark with 98.5% precision to binary-classify 300 million urls.

Dallas, Tx, Us

• Predicted customer churn with web click logs using logistic regression and gradient boosting tree.• Analyzed mobile data usage pattern and used them to predict demographics. • Identified influencers from communication social network of hundreds of millions of users and billions of edges using pyspark, python and various graph analysis tools (graphX, SNAP etc). • Extracted patterns for customer call reasons using association rule mining.• Generated social tags using tens of billions of tweets for 17000 TV shows at scale with pyspark; analyzed correlation between social activities and TV viewership; built topic graphs for TV shows.• Initialized bottom-up data science projects (include expert finder, employee timeline, project recommender and internal news engine), got approved by VP and led a team to implement them using various NLP techniques including topic modeling, keywords extraction and text summarization. • Supervised two interns to perform community detection in large scale social network and implement news recommendations using LDA and word2vec.

Golden, Co, Us

• High throughput computation and data mining on thermoelectric materials.• Predictive cohesive energies using machine learning. I used various algorithms including kernel ridge regression, support vector regression and k-nearest neighbor, etc. to predict cohesive energies of solids on a dataset of over 60,000 calculations and ~9000 materials. The prediction error is reduced to 85 meV/atom, twice lower than previous record.

San Francisco, Ca, Us

• Created newsline.me, a web app to provide context for news in the form of a timeline.• Utilized Google and New York Times APIs to retrieve news articles in real time.• Filtered urls, tokenized and removed stopwords from the news, and stored the data in MySQL.• Applied machine learning (unsupervised clustering) with scikit-learn using TF-IDF (term-frequency inverse document frequency) and n-grams algorithms.• Visualized the timeline and the feature space of the unsupervised clustering using jQuery and D3.• Deployed an interactive web app using Flask, Twitter Bootstrap. Hosted the app on AWS.

Stanford, Ca, Us

• Designed an automated pipeline using Python, Bash and Matplotlib to build structural models, submit simulations to supercomputers, collect data, visualize, and analyze results with regression models.• Mined semi-structured numerical data involving electronic and magnetic properties of metal oxides and hydroxides using Python and SQLite, analyzed the trend using d-band model and identified correlations between different structures.• Improved the simulation time for chemical reactions of molecules on surfaces 40x by porting Python/C code to GPUs : https://trac.fysik.dtu.dk/projects/gpaw/browser/branches/rpa-gpu-expt• Diagnosed and resolved system wide intermittent crashes in collaboration with interdisciplinaryengineers.

Research Associate, Slac

Doctor Of Philosophy (Phd), Physics

Bachelor'S Degree, Physics