Zerodha runs one of India's largest retail brokerages on a deliberately lean stack, so its system-design rounds prize low latency and correctness over web-scale CRUD.
Expect problems anchored in real broking constraints: acknowledging a Kite order in single-digit milliseconds during the 9:15 AM open surge, fanning out live market ticks to millions of concurrent users, and reconciling trades with the exchange and depository so no position is ever wrong. Interviewers reward candidates who treat a reconciliation bug as a regulatory event and design for idempotency, audit trails, and clear mismatch handling.
Recruitment process
Resume shortlist or coding assessment
Technical coding round on DSA and clean implementation
System-design round on a low-latency, correctness-critical service
Founder or senior-engineer round on depth and fundamentals, then offer
Interview rounds
Round 1 (60 min): coding round on data structures, algorithms, and clean code
Round 2 (60 min): system design of a low-latency broking service with correctness constraints
Round 3 (45-60 min): deep-dive on fundamentals, trade-offs, and reconciliation edge cases
Latency and correctness framework
Identify the hot path and its latency budget first, then design for correctness with idempotency and reconciliation, and be explicit about what happens on a mismatch
Practice questions
01
We have 1M users hitting the order-book endpoint simultaneously at 9:15am. We don't want to use Redis or any managed cache. How would you handle it?
IntenseZerodha intentionally constrains stack to test fundamentals. In-process cache, careful invalidation, single-binary thinking. 'I'd use Redis' fails the round.
Design the order-matching flow for Kite so a market order gets acknowledged in single-digit milliseconds during the 9:15 AM open surge. Where are your bottlenecks?
IntenseZerodha runs India's largest retail brokerage on a lean stack; system-design rounds probe low-latency, market-open spikes, and correctness under load rather than web-scale CRUD.
How would you build the end-of-day reconciliation between trades, the exchange, and the depository so no position is ever wrong? What happens on a mismatch?
IntenseCorrectness-over-cleverness probe. In broking, a reconciliation bug is a regulatory event; interviewers want idempotency, audit trails, and a clear mismatch-handling policy.
The typical Zerodha recruitment process has 4 stages: Resume shortlist or coding assessment → Technical coding round on DSA and clean implementation → System-design round on a low-latency, correctness-critical service → Founder or senior-engineer round on depth and fundamentals, then offer.
Zerodha typically conducts 3 interview rounds: Round 1 (60 min): coding round on data structures, algorithms, and clean code; Round 2 (60 min): system design of a low-latency broking service with correctness constraints; Round 3 (45-60 min): deep-dive on fundamentals, trade-offs, and reconciliation edge cases.
HireStepX recommends the Latency and correctness framework for this type of interview: Identify the hot path and its latency budget first, then design for correctness with idempotency and reconciliation, and be explicit about what happens on a mismatch.
Zerodha intentionally constrains stack to test fundamentals. In-process cache, careful invalidation, single-binary thinking. 'I'd use Redis' fails the round.
Zerodha runs India's largest retail brokerage on a lean stack; system-design rounds probe low-latency, market-open spikes, and correctness under load rather than web-scale CRUD.
Correctness-over-cleverness probe. In broking, a reconciliation bug is a regulatory event; interviewers want idempotency, audit trails, and a clear mismatch-handling policy.
Order acknowledgement and matching under market-open spikes, streaming live market ticks to millions of users, and end-of-day reconciliation between trades, the exchange, and the depository.
Both, but correctness leads. In broking a wrong position is a regulatory event, so interviewers look for idempotency, audit trails, and a clear policy for handling mismatches.