Ryan Jiang Email & Phone Number

San Jose, California, United States

• Developed a YOLOv3 model using TensorFlow and OpenCV to detect buildings in Google Maps street view images, enhancing building height estimation accuracy by 68%• Constructed a multimodal deep learning model to identify construction types by integrating numerical building features and ResNet-152 embeddings derived from YOLOv3-detected building images; Obtained 76% accuracy in construction type prediction and reduced property risk assessment time by 80%; Assisted the development team in deploying the model on the online platform• Leveraged HTML5, CSS3, JavaScript, and Segment Anything Model for the front-end, and the RESTful FastAPI framework for the back-end, to develop and deploy a web platform prototype; Empowered clients to select and interact with their insured property via street-view and satellite imagery, enabling real-time prediction of key property attributes such as construction type, building height and square footage• Retrieved and engineered features from property records database using Hive and Python; Developed an XGBoost model with hyperparameter tuning techniques to improve property valuation• Leveraged NLP techniques to clean clients’ addresses, created 30 geographical heat maps on AWS SageMaker to visualize clients’ location distribution and suggested more effective marketing campaigns to increase customer base• Performed in-depth analysis of large-scale public healthcare organization datasets, including essential information, financial portfolios, and patient satisfaction surveys; Created dynamic dashboards to streamline decision-making for healthcare insurance brokers

New York, United States

• Established a training data pipeline using Rasterio and GeoPandas in Python to systematically collect satellite imageries with its ground truth building footprints• Built a U-Net architecture CNN model using PyTorch to automate extraction of building footprints, square footage, and neighborhood density, achieving an F1-score of 0.83 and laying groundwork for future geospatial applications

Sunnyvale, California, United States

Corporate Capstone Project in collaboration with Volvo's Autonomous Drive & AI team• Developed a music recommender system prototype using Neural Collaborative Filtering and DeepFM models in Python; Integrated user demographics, interaction histories, song metadata, and real-time outdoor scene embeddings to personalize Top 10 song recommendations for drivers, increasing Mean Average Precision from 31% to 66%• Implemented the maximum marginal relevance algorithm to re-rank recommended songs for striking a balance between relevance and diversity, leading to a 38% improvement in user satisfaction metrics validated through A/B testing• Addressed new user cold-start problem by recommending trending local songs aligned with preferences of similar existing users based on cosine similarity in user embeddings

Shanghai, China

• Contributed to a digital transformation project for a Fortune 30 automobile company by conducting 5 case studies on test drive procedure innovation, membership community operations and customer data platforms• Completed a 29-page deck on membership management and facilitated 3 workshops to identify and address current pain points• Partnered with a German luxury car brand to develop its official app using agile methodologies and produced 4 reports on E-commerce platform features, including bundle discount sales campaigns and product catalog design

Shanghai, China

• Built ETL pipeline to automate an interactive auto insurance market performance dashboard that displayed the trend of 32 metrics such as average written premium YoY using R and ggplot, reducing cycle time by 85%• Proactively collaborated with Business Operation and IT team to brainstorm and engineer 50+ features from 8 databases on 1M+ records using SAS and Hive SQL queries; Developed logistic regression and random forest model in Python to personalize promotion strategy, which realized a 12% growth in retention rate based on A/B test• Leveraged NLP techniques with NLTK to clean customers’ addresses, created 6 geographical heat maps to visualize customers’ location distribution and suggested more effective marketing campaigns to increase customer base• Created 3 Tableau dashboards to visualize conversion and retention rates, enabling actuarial managers to refine and improve pricing strategies• Analyzed industry trends and market performance data to produce comprehensive weekly actuarial reports for the Management Committee

Toronto, Ontario, Canada

• Developed an R package to implement approximate Bayesian inference using the Integrated Nested Laplace Approximation (INLA) method for case crossover models• Wrote a 25-page vignette in R Markdown comparing our Bayesian methods with conditional logistic regression• Implemented a scalable modeling framework with linear and semiparametric effects to analyze non-linear associations between mortality rates and extreme temperatures, utilizing over 50,000 records from England (1993-2006)• Conducted research on optimization algorithms, including line search methods (IPOPT) and trust region approaches, to optimize the mode of the log posterior density.

Toronto, Ontario, Canada

• Performed data analysis and visualization over 80 cities’ KPIs in Python and presented corresponding findings to 30+ stakeholders in a clear manner to facilitate understanding and actionable next steps• Wrote policy briefs, working papers, City KPI data collection rules and computing formulas, which were finally published in ISO 37122:2019 - Indicators for Smart Cities

Toronto, Ontario, Canada

• Provided expert statistical consulting services to non-statistics major students at the University of Toronto and collaborated with Human Biology researchers to investigate whether 6 to 8-year-old children could learn control-of-variables strategies through specific teaching methods• Built a logistic regression model with bidirectional stepwise selection for variable selection and utilized Fisher’s Exact Test to evaluate the statistical significance of differences in correct answer rates between experimental and control groups• Identified common characteristics of successful cases and offered recommendations for improving the experimental methodology

Master'S Degree, Industrial Engineering And Operations Research, 3.9/4.0

• Coursework: Machine Learning, Applied Natural Language Processing, Computer Vision, Database Design & Analysis, Data Science Principles.

Honours Bachelor Of Science, Statistics And Economics, High Distinction

• Coursework: Data Structures and Object-Oriented Programming, Probability, Linear Algebra, Multivariable Calculus, Multivariate.

International Baccalaureate Diploma