Startup · Bengaluru
InMobi interviews target mobile advertising and AdTech, covering ML systems, real-time bidding, and large-scale data engineering.
Process: Online Coding → Technical × 2 → System Design → HR
What does InMobi do and what engineering problems are unique to mobile advertising?
Tip: InMobi: mobile advertising and marketing cloud. Core products: DSP (Demand Side Platform), SSP (Supply Side Platform), and Glance (lock screen content). Engineering challenges: real-time bidding (<100ms), petabyte-scale ad event processing, user identity graphs across devices, and ML for CTR prediction on 1B+ ad requests/day.
How does Real-Time Bidding (RTB) work in programmatic advertising?
Tip: RTB flow: User opens app → SSP sends bid request (user context, ad slot) to all DSPs → each DSP scores opportunity and responds with a bid (<100ms total) → SSP runs first-price/second-price auction → winning DSP delivers the creative. InMobi participates as both DSP and SSP — they have full-stack ad tech.
Design a CTR (Click-Through Rate) prediction system for mobile ads.
Tip: Features: user demographics, device type, time of day, ad creative type, publisher context, historical CTR of the ad-user combination. Model: logistic regression (fast, interpretable) or gradient boosting (GBDT) for offline training. Serving: pre-computed user embeddings + online feature assembly. <5ms per prediction SLA.
What is a user identity graph and why is it important for ad targeting?
Tip: Identity graph: connects different device/user identifiers (IDFA, GAID, email hash, cookie) that belong to the same person. Enables: cross-device targeting (show ad on mobile to user who browsed on desktop). Building: deterministic linking (same login), probabilistic linking (similar behaviour patterns). Privacy: India's DPDP Act restricts how this can be done.
How would you build a low-latency ad serving system that handles 1 million requests per second?
Tip: Architecture: stateless ad server (horizontal scale) + pre-computed candidate ads in Redis + in-memory CTR model. Steps: parse bid request → lookup user profile (Redis, <1ms) → fetch candidate ads (Redis sorted set) → score each ad (in-memory model) → return top-K. No DB calls on critical path. Cache miss: async profile lookup, default to demographic targeting.
Write code to implement a rate limiter using a sliding window algorithm.
Tip: Sliding window log: maintain a sorted set of request timestamps. On each request: remove timestamps older than window, check count, add current timestamp. Redis ZSET: ZREMRANGEBYSCORE + ZCARD + ZADD in a Lua script for atomicity. O(log n) per request. More accurate than fixed window (no boundary spike).
Explain how ad frequency capping works and implement it.
Tip: Frequency cap: limit how many times a user sees the same ad (e.g. max 3 times/day). Implementation: Redis counter per (user_id, campaign_id, date). INCR: increment on each impression, EXPIRE at midnight. Check: if counter >= cap, exclude ad from auction. InMobi enforces frequency caps globally across millions of users.
What is Apache Spark and how would InMobi use it for ad event analytics?
Tip: Spark: distributed processing engine for large-scale data. InMobi uses Spark to process: daily ad event logs (impressions, clicks, conversions → 100TB+), compute campaign performance metrics, retrain CTR models. Key: Spark SQL for analytics, MLlib for model training at scale. Compared to MapReduce: 100× faster with in-memory DAG execution.
Tell me about a time you optimised an algorithm or system to meet a strict latency SLA.
Tip: InMobi's RTB must respond in <100ms (IAB standard). Walk through: what the SLA was, your profiling process (flame graphs, distributed tracing), what you found (top bottleneck), the optimisation (algorithm change, cache, parallelism), and the measured improvement. Numbers are essential.
What privacy regulations affect mobile advertising and how does InMobi adapt?
Tip: Key regulations: GDPR (EU), CCPA (California), Apple ATT (iOS 14+ — IDFA opt-in required), Google Privacy Sandbox (replacing GAID). InMobi's response: contextual targeting (app content, time) instead of user-level, cohort-based targeting, first-party data partnerships. India's DPDP Act (2023) adds further requirements for consent management.
InMobi Pte. Ltd. interviews follow a 4-round process. Here is what to expect and how to prepare for each stage.
Upload your resume and get questions scored across technical depth, communication, structure, confidence, and relevance — the same criteria InMobi panels use.