Personalized Learning Recommendation Platform

Lifting course completion from 34% to a 55%+ benchmark with hybrid AI personalization.

Role

Product Lead (Self-driven)

Timeline

2025

Completion lift

+30%

Engagement

+25%

Problem

34% course completion vs. a 55% industry benchmark, with “irrelevant content” ranked the #1 drop-off reason across exit surveys.

Solution

Designed a modular-monolith recommendation platform on AWS Personalize evolving into a custom ML stack — a personalized feed, AI-generated learning paths, re-engagement nudges, tiered consent, and admin-side A/B testing. Hybrid batch + real-time inference delivered at ~20% of pure real-time cost, with a dedicated ‘explore’ slot to prevent filter bubbles.

My contributions

• Authored 18 functional requirements across 3 personas (learner, instructor, admin).
• Defined the inference architecture and cost model with engineering.
• Set North-Star and guardrail metrics for relevance, diversity, and trust.
• Built a phased experimentation plan with explicit success and rollback criteria.

Results & learnings

• Target +30% course completion, +25% engagement, 15–20% NRR uplift.
• Cost-per-recommendation reduced ~80% vs. all-real-time baseline.

Stack & methods