I'm Jian Xu.

Courses: Deep Generative Models, Web Applications, NLP, Machine Learning, Probability Analysis, Data Analysis in Physical Systems

Courses: Algorithm Design & Analysis, Operating Systems, Computer Networks, Graph Theory & Algorithms, Advanced Programming, Computer Graphics, Digital Image Processing

Courses: The C Programming Language, Data Science and Innovation, Engineering Mechanics, Fluid Mechanics, 3D Printing, Mathematical modeling in Computer Science

• Designed AI-based solutions to identify fraudulent activities in sample order requests on Renesas' E-commerce platform.
• Implemented a data collection pipeline on AWS, using Lambda Functions, triggered by data uploads to an S3 bucket. This process efficiently dispatched API fetch requests to SQS, with the resultant data stored in DocumentDB.
• Conducted data preprocessing, data augmentation, feature engineering, and trained weighted-ensemble ML models for fraud detection, achieving a 95% accuracy post-production.
• Engineered a serverless cloud workflow on AWS, integrating API Gateway and Lambda functions to create a RESTful backend API, effectively managing order request processing.
• Deployed SageMaker endpoints to detect fraudulence for real-time fraud detection in order requests, which prevented 60% financial loss, reduced 50% labor cost, and decreased the overall order request processing time by 50%.

• Addressed the lack of topological directions in GNN, a key factor in over-smoothing and low expressiveness.
• Modeled graph direction with vector field and designed asymmetric aggregation inspired by CNN anisotropic kernels.
• Implemented directional message-passing aggregators based on Laplacian matrix gradient using Pytorch and DGL.
• Evaluated the model in various GNN tasks, with an average 3% increase in performance and potential interpretability.

• Fine-tuned an OCR model specifically for invoice image using PaddlePaddle, with an accuracy increase of 15%.
• Implemented interactive bounding boxes using React on invoice images, enabling dragging and resizing adjustments around target area. This process segments the image into smaller sections, based on the bounding boxes' coordinates.
• Developed RESTful APIs using Django for image processing, formatting OCR model outputs, and saving to MySQL databases. Successfully onboarded 4 hospitals to the semi-automatic invoice recognition tool, achieving a 75% reduction in manual invoice input time.