PhonePe Interview Questions & Answers 2026

Architecture: API Gateway, Auth service, Payment Orchestrator, NPCI integration, Bank routing, ledger service, notification, fraud detection. Handle scale: 12+ billion transactions/year, peak 50K+ TPS. Critical: idempotency, eventually consistent reconciliation, high availability (99.99%), low latency (<2 sec end-to-end).

Client generates UUID (idempotency key) per transaction. PhonePe stores key with response in Redis (24-48hr TTL). On retry: check key - if exists, return cached response; else execute and cache. Database also has unique constraint on (user_id, idempotency_key) as backup. Critical for at-least-once delivery semantics in payments.

For real-time TPS monitoring across regions. Use sharded counters with periodic aggregation. Each region writes to local Redis. Aggregator reads all regions every 1s, computes total. Use HyperLogLog for unique counts. For exact counts at scale, use Kafka stream processing with windowing.

Classes: User, Group, Expense, ExpenseSplit (equal/exact/percentage). Operations: createExpense, settleUp, getBalances. Algorithm: simplify debts using graph (minimize transactions). Handle: groups, currency conversion, recurring expenses. 90 min round, focus on OOP design and edge cases.

Approach 1: Expand around center - for each i, check odd-length palindromes (center i) and even-length (center i, i+1). Time O(n^2), Space O(1). Approach 2: Manacher's algorithm O(n) but complex. Approach 3: DP O(n^2) time and space. Use approach 1 in interview, mention Manacher.

PhonePe is India's leading UPI payments platform, processing ~46% of all UPI transactions. Independently listed/spun off from Walmart-Flipkart group. Mention their scale (500M+ users), tech innovations (PhonePe Switch, ATMs, POS), expansion into insurance, mutual funds, gold. Strong engineering culture with focus on reliability.

1) Identify if it's actually slow via metrics (P50, P95, P99). 2) Check distributed tracing (Jaeger/Zipkin) for slow spans. 3) Database query analysis (EXPLAIN, slow query log). 4) Cache hit rates. 5) External API dependencies. 6) Network latency. 7) Resource saturation (CPU, memory, connections). 8) Fix: index, cache, async, batch. Verify with load test.

Use STAR method. Describe baseline metric (e.g., P99 latency 2sec), profiling approach (APM, custom metrics), bottleneck identified, optimization implemented (caching, query optimization, code refactoring), measurable improvement (latency reduced 60%), and rollout strategy.