From Generic to Contextual: Personalization Redesign Using LLMs at an EdTech Startup

AI · 6 min read

The startup’s old recommender used static rules based on completion percentages and explicit tags. That led to repetitive or irrelevant recommendations, particularly for learners with irregular schedules. The product team experimented with an LLM that synthesized recent activity, stated goals, and free-text progress notes to produce short, contextual prompts like “Try a 10-minute micro-lesson to reinforce last week’s concept.”

Design constraints prioritized transparency and user control: recommendations displayed why they were suggested, offered a one-tap “Not right now” option, and allowed users to tune the recommendation frequency. The UI used compact cards with clear CTAs and a lightweight feedback mechanism that fed back into the model for retraining.

During a 10-week rollout, click-through to recommended lessons rose 24% and weekly active learners increased 9%. Qualitative feedback showed learners appreciated shorter, context-aware suggestions rather than generic course lists. The project highlighted that personalization powered by LLMs needs simple UX levers for trust and an explicit feedback funnel to correct drift.