Onsite System Design: App User Behavior Data Platform

Design a system to analyze Uber App user behavior data and process raw events into downstream-ready data for models and business systems. The key is not just collecting clicks, but building pipelines for sessionization, funnel analysis, feature generation, quality checks, and schema evolution. Requirements Collect app behavior events such as clicks, sessions, conversions, refunds, ratings, and order-completion outcomes. Support funnel analysis that links user actions across the full journey, not isolated click counters. Produce processed data assets usable by downstream models, analytics dashboards, and business systems. Support both historical queries and fresher pipeline outputs where needed. Define event APIs, schemas, data model, ingestion path, cleaning, aggregation, storage, and serving surfaces. Handle schema evolution, event-tracking quality, late / duplicate events, and data-quality monitoring. Notes The prompt starts deliberately broad; the interviewer may answer clarification questions with little direction. Drive the round by proposing concrete user-behavior examples and asking which downstream consumer matters most. A strong answer moves beyond a generic analytics dashboard into data processing for model users: sessionization, funnel features, offline / realtime pipelines, and reusable feature tables. The interviewer signaled that simply describing collection and dashboarding is only the starting point. Preparation Prepare a canonical event schema with user id, session id, event name, entity ids, timestamp, client metadata, and version. Practice a dual-path architecture: streaming aggregation for fresh signals plus batch backfill for correctness. Drill data-quality checks: duplicate events, missing IDs, schema version mismatches, late arrivals, and metric reconciliation.