MLSD — Nearby Place Recommendation

Location-aware recommender for nearby places (Marketplace shops, FB Places). Geohash retrieval + mobile real-time constraints + context features distinguish it from generic feed MLSD. Requirements Functional: given user location + intent (e.g., browsing Places near me), return ranked list of places. Retrieval: geohash / quadtree spatial index → candidates within radius; combine with content embedding retrieval for personalization. Ranking: standard 4-stage scaffold (retrieval → pre-rank → rank → re-rank) layered on top of the spatial pre-filter. Features: distance, dwell-time history, hour-of-day, device, friends-who-visited graph signal. Realtime constraint: user is mobile — location updates frequently; refresh interval and stale-cache trade-off. Metrics: click-through, visit conversion, complaint rate. Notes Multiple reports stress: interviewer probes metric design and feature engineering more than architecture. Lead with metrics + features, not model choice. Mobility / freshness is the differentiator from a static Reels recommender — highlight it early. Preparation Memorize geohash vs quadtree trade-offs (static vs dynamic locations). Practice the metric + feature pitch first; only then dive into model. Prep an A/B-test design where mobile-session length is the key randomization unit.