Tsunghan (John) Wu

AI Infra - Build failure debugging agents for ML training jobs / infra (e.g. GPU/CPU OOM, stuck jobs, NCCL watchdog timeout) - Performance optimization for ML training jobs / infra Ads Ranking, ML Infra - Enable SOTA architectures for Ads ranking models - Enable Transfer Learning for Ads ranking models - ML Model Platform reliability

- News ranking and recommendation. - Push notification optimization.

Bridged machine learning, deep learning, and AI capabilities to automate various business process operations by applying cloud services, ML/DL models, transfer learning, and in-house AI solutions. - Built B2B product recommendation engines (memory-based collaborative filtering, model-based collaborative filtering (with WMF / ANN), and content-based models) for precision marketing. - Built GraphQL APIs to coordinate ML backend APIs, React Web App, Django RESTful APIs, and Postgres database for the revenue growth system. - Designed and implemented AI-driven document processing automation by leveraging computer vision, natural language processing/understanding, machine learning, deep learning, and knowledge representation techniques. - Built a platform to benchmark in-house information extraction models for processing real-world documents. - Applied ML and NLP for various content moderation processes, including both text analysis and image comprehension. - Built and Deployed websites for AI demos and REST APIs by using Django and Flask frameworks. - Built web crawlers and conducted in-depth data analysis to help clients discover potential customers, enhance marketing efficiency, and increase business growth. - Implemented a database reconciliation automation program for a large utility company.

- Framed an auxiliary Osteoporosis diagnosis system which automates the interpretation of Bone Mass Density reports. The system helps orthopaedists speed the diagnosis and reporting process. - Analyzed meteorological and environmental data for Red Quinoa and built a predictive model for forecasting crop production and pests, which helps farmers enhance the quality and quantity of their crop.

Designed, prototyped and implemented an end-to-end question answering system for solving Accounting computational questions which leveraged machine learning, information extraction, natural language processing (NLP), and Neo4j graph database based on a novel cognitive computing framework.

Dissertation: Tracking Dynamics of Temporal Social Networks and Applications in Structural Network Analysis. Keywords: temporal bipartite networks, temporal social networks, PageRank, spectral graph theory, Laplacian eigenmaps, finite impulse response filter, recurrent neural network, community detection (clustering), link prediction, node ranking.

Thesis: Implementation of BB84 and SARG Protocol on Quantum Key Distribution System.