Track: Recursive Harnessing. SwipeHome is a self-improving real-estate recommendation harness built on MongoDB Atlas. Before showing a home, it predicts whether the user will like it and stores that prediction. User actions become measurable feedback that updates persistent preference memory across 36 home traits. After enough interactions, SwipeHome analyzes its prediction failures, proposes constrained changes to its own retrieval, ranking, memory, and exploration policies, and backtests those changes on newer holdout interactions. A new policy is promoted only when it performs better; failed policies are rejected and preserved. MongoDB Atlas stores properties, predictions, interactions, memories, experiments, and policy versions, while Atlas Vector Search powers personalized retrieval. SwipeHome learns not only what homes a user likes, but how its recommendation system should learn that user.
Team
Placement
Participant
