Research assistant at the University of British Columbia, working on the intersection of reinforcement learning and persuasive system design.
Research Focus
The core question: can we build software systems that encourage lasting positive behavior change — not through manipulation, but through intelligent timing and framing of interventions?
Methodology
- Modeling user behavior as Markov Decision Processes using click-stream data
- Training RL agents that optimize for long-term behavior persistence (90+ day horizons)
- Building recommendation models that adapt intervention strategies based on individual user response patterns
Key Findings
Preliminary results show a 2.3x improvement in 90-day behavior persistence compared to rule-based intervention systems. The agent learns to back off when users show fatigue and vary intervention types to prevent habituation.
Ethical Framework
All research operates under strict ethical guidelines — agents cannot optimize for engagement metrics, only for outcomes the user has explicitly consented to.