Yawen Zhao

AI Engineer at Mozat Singapore

Designed and implemented a graph-driven multi-agent conversational system integrating intention graphs, probabilistic transition matrices, and role-specific speaker modules (friendship, love, psychotherapy). Achieved personalized conversation flow through real-time user profiling and intention graph matrix transformations, reinforcement learning from user feedback, and memory-augmented decision making to maintain coherence while adapting to user preferences in real time.

AI Engineer at Mozat Singapore

Designed and implemented a graph-driven multi-agent conversational system integrating intention graphs, probabilistic transition matrices, and role-specific speaker modules (friendship, romance, therapy) to support real-time user profiling and intent-graph transformations. Implemented adaptive learning components (memory-augmented decisions) and a dynamic graph that updates transition probabilities based on conversation performance, achieving natural topic continuity and personalized user experiences.

AI Engineer at Mozat Singapore

Designed and implemented a graph-driven, multi-agent conversational system with memory-augmented learning, real-time user profiling and feedback-based adaptation; integrated intention graphs and speaker modules for friend/romance/psychotherapy contexts, achieving personalized conversational flows and high engagement.

Research Intern at Nanyang Technological University

Node-Time Conditional Prompt Learning for Financial Fraud Detection on Dynamic Graphs. Implemented a novel Temporal Graph Network (TGN) framework with dual-prompt learning, enabling parameter-efficient adaptation from self-supervised pre-training to downstream dynamic graph tasks.

Research Intern at Academic Research

Participated in node-time conditional prompt learning for financial fraud detection on dynamic graphs; implemented Temporal Graph Network framework; developed node-time conditional prompts using condition-net and prompt-tuning across data splits, enabling robust fraud detection across dynamic graphs.

Research Assistant at Nanyang Technological University

Comparative Analysis of Sequence-to-Sequence Architectures for English-German Neural Machine Translation (Phase 2). Extended experiments across additional data splits and model variants to validate robustness of MT architectures.

Research Assistant at Nanyang Technological University

Comparative Analysis of Sequence-to-Sequence Architectures for English-German Neural Machine Translation. Implemented multiple Seq2Seq translation models for a large English-German dataset; compared LSTM with attention, standard LSTM, GRU with attention, bidirectional LSTM, and Transformer architectures.

Data Analyst Intern at Hive Analytics

Ingested raw NYC taxi trip data into Hadoop HDFS and set up Hive tables to run distributed SQL queries, verify data quality, and perform initial aggregations on trip distance, time, and fare fields.

Market Insights Intern at IQVIA

Market Insights Intern (Data Analytics Track). Conducted descriptive statistics on health care related datasets; performed market research on medical technology companies and medical devices for client projects; compiled relevant literature, industry information, and news to support strategic decisions.

Market Insights Intern (Data Analytics Track) at IQVIA

Conducted market research on medical technology and medical devices for healthcare clients; compiled literature, industry information, and news to support strategic decisions.

Data Analysis Intern at Alibaba Hangzhou

Conducted descriptive statistical analyses on home-stay bookings, evaluating indicators such as customer loyalty, reservation preferences, off-peak seasons, and matching rates to inform business decisions and product improvements.