Xinhui Kang

Technical training in red-teaming, mechanistic interpretability, and alignment/safety evaluation with a mentored capstone evaluation research project on AI safety

MARS 4.0 Programme Project: Inspect Evals – AI R&D; developing standardised open-source AI R&D benchmarks in the Inspect framework for capability monitoring. Refactoring and hardening existing AI R&D evaluations with testing and bug-fixing, and co- authoring a paper on methodology and best practices.

Quantifying LLM-advice–induced polarization on controversial issues, with sycophancy and party- alignment benchmarks, moderator modeling (ideology/decision-style/openness/AI-trust), and cross-model comparisons vs. human and prior PNAS baselines.

LLM agent on egocentric video understanding for AR-Based applications Computational framework for large-scale semantic annotation of paintings

Build batch and streaming data pipelines using Spark and Flink in Python, Scala and Java Generate data assets for analytical operations such as user profile analysis and revenue reconciliation

• Perform data analytics using SQL on Google BigQuery • Generate dashboard visualisation using Google Data Studio • Perform machine learning and natural language processing tasks on chat log messages in Python • Build Chatbot for automated response using Google Dialogflow and perform integration with Google Sheet API, Google

• Conduct UI/UX studies • Conduct user testing on robot management application. • Perform Human Computer Interaction studies to optimise user interface

• Obtained Automation Anywhere Advanced Certificate • Automated the process of collecting KPI variables for the project management team and optimised the data collection process using Automation Anywhere • Design Tableau Dashboard for KPI reporting • Researched on tools such as Automation Anywhere and Power Automate to optimise bot run efficiency • Deployed script to production to automatically retrieve control room data and inputs from business users to consolidate report for IT product manager • Conduct bot testing with business users to align the business requirement and technicality of the scripts

• Perform data analysis on client data from Cards and Loans Servicing team • Implement improvement on the existing operational practices (for e.g. improving the process of issuing cashier's order, and managing undelivered returned credit cards) • Liaise with offshore IT team in India and Sri Lanka to resolve technical issues, and communicate the demand of my team to the offshore IT team • Assist to bridge communication between the HSBC Operations team and the NETS team, and visited NETS office to perform auditing activities like stock check-up • Participate in weekly Operations department meeting with office in Hong Kong to summarise weekly update for manager to refer