From Activity to Intent: Generating User Journeys with LLMs

Back to Devoured - October 12, 2026

DEVOURED

From Activity to Intent: Generating User Journeys with LLMs

Pinterest replaced brittle keyword-clustering with a distilled 4B parameter LLM to map user activity to high-intent 'journeys' at scale.

What: Pinterest transitioned to a fine-tuned, smaller LLM that converts raw user activity into structured user goals. The model uses teacher-model distillation to maintain performance while keeping operational costs feasible.

Why it matters: This represents the 'small model' shift: using highly specialized, distilled models for specific high-frequency classification tasks rather than relying on massive general-purpose models.

Decoder

Distillation: A process where a smaller 'student' model is trained to mimic the behavior and outputs of a larger, more powerful 'teacher' model.

Original article

Pinterest replaces a fragile keyword-clustering pipeline with a fine-tuned 4B LLM that converts a user's activity history into ranked, localized “journeys” (active goals). Teacher-model distillation makes it affordable at scale. The system improves notification engagement while applying strict output validation, safety filters, and feedback loops.

Explore all 71 articles in today's digest