Argued that 'human-AI collaboration' is a misleading metaphor unfair to data workers in the Global South
Demonstrated that ranking (vs. scoring) is faster and more accurate for capturing human judgment—applied in multiple sclerosis assessment and word complexity
Designed a visualization tool that drastically lowers expertise barriers for large time-series data analysis; won multiple awards and is open-source
Showed error bars can function as control mechanisms, not just uncertainty indicators
Studied how data 'sketching' facilitates analyst conversations and eye-only data navigation
Found public reluctance toward self-driving cars, low enjoyment in chatbot therapy, and limitations of natural language for voice assistant interaction
Proposed rule-based specification (over like/dislike feedback) for programming intelligent systems like recommenders
Research Experience
Researcher at Microsoft
Affiliated Lecturer at the University of Cambridge
Honorary Lecturer at University College London
Expands spreadsheet technology to build machine learning models and compose/perform music
Investigates viewing spreadsheet formulas as textual programs to improve efficiency and reduce errors
Studies user-friendly reuse of spreadsheet grids, integration of Python code in spreadsheets, spreadsheet-based database transformation scripting, and unit-aware spreadsheets (e.g., meters, dollars, kilograms)
Explores human interaction with Generative AI, proposing AI as a 'provocateur' to enhance critical thinking and intentionality
Developed a working definition for Generative AI to clarify the term
Created visualization tools for analyzing fault patterns in the British Telecom network