PhonePe Interview Questions 2026

Tip: PhonePe processes over 6 billion transactions per month — one of the largest UPI volumes globally. Stack: Java/Go microservices, Kafka for event streaming, MySQL + Cassandra, in-house ML for fraud and recommendations. Engineering interviews emphasise scale, reliability, and fintech domain knowledge.

Tip: Fan-out: Kafka topic per notification type → consumer workers → FCM/APNs. Throughput: partition Kafka by user_id for ordering guarantees. Rate limiting per device to prevent spam. Priority queues (transactional > promotional). Dead letter for failed deliveries. Track: delivered, opened, converted.

Tip: UPI: NPCI-governed protocol on top of IMPS. Flow: user initiates transfer → PSP (PhonePe) encrypts with UPI PIN → NPCI authenticates → source bank debits → destination bank credits → confirmation. Key: PIN never leaves device (end-to-end encrypted). PhonePe interviews expect this flow knowledge.

Tip: Feature signals: transaction velocity (>10 txns/min), new beneficiary + high amount, SIM-swap in last 24h, device change. Rule engine (<5ms) for blocking obvious patterns. ML model (gradient boosting) for probabilistic scoring. Human review queue for borderline. RBI mandates: block if >3 failed PIN attempts.

Tip: Offline: use a HashMap + min-heap of size K. O(n log k) time. Online/streaming: Count-Min Sketch (probabilistic, O(1) space per element) + heap for top-K. PhonePe data interviews often involve streaming aggregation — know both exact and approximate algorithms.

Tip: Account balance is eventually consistent between PhonePe (PSP) and the actual bank. PhonePe maintains its own ledger that reconciles with bank statements every few minutes. The displayed balance is from the PhonePe ledger, which may lag bank reality by seconds. Design: event sourcing the balance change log for auditability.

Tip: In fintech, wrong > slow is worse than slow > wrong. Show you understand this: describe a time you pushed back on a tight deadline for a financial feature because correctness could not be compromised, and how you quantified the risk of shipping prematurely.

Tip: Synchronous: user waits while the entire bank debit + credit happens — slow (banks may take 30s). Asynchronous: PhonePe returns "transaction initiated" immediately, processes in background, notifies via push/SMS when confirmed. UPI mandates 30s timeout — PhonePe must handle pending states gracefully.

Tip: Pre-materialise: write transaction events to a user-specific feed table (append-only, Cassandra) in addition to the canonical ledger. Page fetch: single Cassandra read for the latest 20 rows. Infinite scroll: cursor-based pagination (no OFFSET). CDN caches icons and category metadata. Client-side: skeleton screens for perceived performance.

Tip: Cassandra: wide-column NoSQL, leaderless replication, tunable consistency (eventual or quorum). Strengths: write-heavy workloads, linear horizontal scaling, no single point of failure. Use over MySQL when: >100K writes/sec, data model is key-value or time-series, no need for complex joins. For transaction logs, event feeds — ideal.