LLMs can reliably abstract noisy, granular user interaction logs into interpretable workflows without domain-specific training, enabling better product insights while preserving privacy.
WorkflowView uses large language models to convert low-level user action logs (clicks, keystrokes, etc.) into high-level, meaningful activities. The approach works across different applications and domains—from browser history to online courses to document editing—without needing task-specific training, making it practical for understanding real user behavior at scale.