Product Company Interview Questions India — Flipkart, Coupang, Razorpay, Microsoft

Most engineers preparing for product company interviews in India study the same LeetCode list, watch the same system design YouTube videos, and then walk into a Flipkart machine coding round with zero compilable code practice — and wonder why they failed. The engineers who crack product company interview questions in India at the SDE-2 and MLE level are not smarter; they prepped for the specific company in front of them, not a generic "top tech company." Coupang's bar is closer to FAANG algo difficulty. Razorpay will probe your understanding of idempotency until you either know it cold or visibly don't. Flipkart will hand you a laptop and expect working code at the end of round two. If you're prepping identically for all three, you're leaving offers on the table.

How India's Top Product Company Interviews Actually Differ

Generic "product company interview questions India" advice treats every company as interchangeable. They are not. Here is what the loop actually looks like at each company, company by company:

The companies on KnowEase's platform — Flipkart, Coupang, Microsoft, Razorpay, Uber — each have mapped interview loops with round-by-round breakdowns. Use them instead of guessing.

Flipkart SDE-2 Interview: What the Loop Actually Looks Like

Flipkart runs 4–5 rounds for SDE-2. Here is the sequence:

Round 1 — Coding Screen (DSA). Medium to hard difficulty. Expect graph traversal, DP, or interval problems. LeetCode medium is not enough — you need hard comfort. Time complexity analysis is mandatory; interviewers probe it directly.

Round 2 — Machine Coding / LLD. This is Flipkart's signature round and where most candidates fail. You get a laptop, an IDE, and 60–90 minutes to implement a working system from scratch. Common problems: design a cab-booking system, design a splitwise application, design a library management system. They run your code. It must compile. It must handle edge cases.

What they actually score you on in this round is not "did it work" — it's how your design handles requirements changes introduced mid-round. An interviewer will say "now add support for premium users with priority booking" 40 minutes in. Your code's response to that addition reveals everything about whether you understand SOLID principles, open-closed design, and proper interface segregation.

Round 3 — High-Level System Design (HLD). Problems that have come up repeatedly: design Flipkart's recommendation engine, design the cart service for a flash sale, design a warehouse management system with real-time inventory. The scale context is Indian e-commerce: Big Billion Days-level traffic spikes, significant rural network variance, and a mix of payment methods including CoD.

Round 4 — Bar-Raiser / Hiring Manager. Behavioral plus a light technical probe. They're testing for ownership and depth of past work, not reciting company values.

For LLD specifically — patterns, SOLID violations, and machine coding practice — the full framework is in Object-Oriented Design Interview Questions 2026. That post covers the exact class design patterns Flipkart and Coupang score you on.

Coupang's Interview Is Closer to FAANG Than Most Indian Product Companies

Coupang runs 5 rounds and the algo difficulty is genuinely near LeetCode Hard. This is the company in India where pure DSA weakness gets exposed fastest.

Round 1 — Online Assessment. 2 problems, 90 minutes, LeetCode Hard caliber. Graph problems with complex constraints are common. If you're LC Medium comfortable and LC Hard uncomfortable, prepare for this to filter you before the loop.

Rounds 2–3 — Technical Interviews. Each is 60 minutes: one algorithm-focused, one system design. The system design round at Coupang is specifically oriented toward ML systems and real-time infrastructure. Rocket Delivery — Coupang's same-day / next-day delivery promise — requires sub-second decisions at the routing layer. Interviewers ask about:

• Real-time ML serving pipelines: how do you serve a model predicting delivery windows at sub-100ms P99 latency?
• Feature store architecture: online vs. offline feature stores, freshness guarantees, feature skew between training and serving
• Demand forecasting system design: how do you predict item-level demand across 50,000 SKUs for a flash sale? What's your retraining cadence?

Round 4 — ML System Design (for MLE roles). This is where Coupang differentiates from other Indian product companies. They expect you to understand the full ML lifecycle in a production context — not just model training. Expect questions about: A/B testing infrastructure, shadow mode deployment, model rollback triggers, and real-time feature pipelines.

Round 5 — Behavioral. Structured, not LP-formal. They probe past projects with depth ("what broke in production?", "what would you do differently?"), not values recitation.

For a deeper look at how Coupang and other companies build ML infrastructure at scale, see How Big Tech AI Systems Work at Amazon, Coupang, and Walmart — it covers the exact architectural patterns Coupang interviews test.

Razorpay and Fintech Interviews: Distributed Systems Are the Filter

Razorpay, PhonePe, and Paytm share an interview DNA: they care deeply about correctness under failure. Surface-level distributed systems knowledge — "use Kafka for async, use Redis for caching, CAP theorem says consistency vs. availability" — fails here. The filter is whether you understand what happens when things go wrong.

Here are specific questions that come up in Razorpay and fintech interviews in India:

1. "Design a payment processing system. How do you guarantee exactly-once semantics when the network call to the payment gateway times out?"

The right answer involves idempotency keys at the API layer, at-least-once delivery with deduplication, and database-level unique constraints. The wrong answer involves hoping the retry doesn't cause a double charge.

2. "We're processing ₹10,000 crore in daily transactions. A database node fails mid-transaction. How do you ensure money is never lost or double-counted?"

This probes two-phase commit awareness, saga pattern for distributed transactions, and outbox pattern for reliable event publishing. Candidates who answer "use distributed transactions" without knowing 2PC's failure modes miss the point.

3. "Explain how you'd implement an idempotent refund API."

The complete answer: idempotency key in the request, store the key + response in a dedicated table before processing, return stored response on duplicate key detection, handle the race condition on concurrent identical requests with row-level locking or SELECT FOR UPDATE.

4. "Your payment service sends an event to the notification service after a successful payment. The notification service is down. How do you ensure the SMS is eventually sent?"

Outbox pattern: write the event to an outbox table in the same transaction as the payment record. A separate process polls the outbox and publishes to the message queue. This guarantees the event is published if and only if the payment committed.

The correctness-under-failure bar at fintech companies is why engineers who only prep for Swiggy-style system design questions get surprised by Razorpay loops. They're testing different things.

Microsoft India: System Design Plus Behavioral Depth

Microsoft India runs 4–5 rounds for SDE-2. Coding difficulty is moderate — medium LeetCode with clean code expectations. The filter is behavioral depth and system design quality, not algorithm complexity.

Microsoft's behavioral round is LP-style in structure but less codified than Amazon. They use competency-based questions ("tell me about a time you disagreed with your manager on a technical decision") and probe for specific examples with outcome metrics. Vague answers fail here even if the story is technically interesting.

System design problems are broad: design a collaborative document editor, design a notification service for Teams, design a distributed cache. They're testing your ability to reason about trade-offs clearly, not whether you know every trade-off in advance. Talk through decisions explicitly. "I chose Postgres over DynamoDB here because our access patterns are relational and we're not at a scale where DynamoDB's cost makes sense" is the right register.

The 3 Things Every Indian Product Company Tests (Regardless of Company)

Across Flipkart, Coupang, Razorpay, Microsoft, Swiggy, Uber — the specific rounds differ, but three patterns appear in every loop. Prep for these and you've covered the universal foundation of product company interview questions in India.

1. The 10x Traffic Inflection Mid-Interview

Every system design round includes this moment: you've sketched your architecture, and the interviewer says "now your traffic just went 10x. What breaks first, and how do you fix it?"

This is not a trick question. It's a test of whether your design has a graceful degradation path. Engineers who answer well do three things: identify the specific bottleneck (usually the database write path, or the synchronous API calls in the critical path), propose a concrete fix (read replicas, write-behind cache, async processing via queue), and acknowledge the new trade-offs introduced. Engineers who answer poorly either freeze or propose a complete redesign — neither of which you have time for in production.

2. LLD Embedded in Coding Rounds

Companies with a dedicated LLD round (Flipkart, Coupang) make it explicit. Companies without one (Swiggy, Razorpay) still test it implicitly — they evaluate the class structure and extensibility of the code you write in the DSA round. An SDE-2 who solves the algorithm correctly but writes a 300-line procedural function gets flagged for "not designing like a senior engineer."

Design your coding round solutions with classes, interfaces where appropriate, and named constants. This signals SDE-2 level design thinking even in a DSA context.

3. Clarifying Questions Under Ambiguity

Every interviewer at every Indian product company — without exception — deliberately gives incomplete requirements. The SDE-2 who immediately starts designing fails this test. The one who asks "what's the expected read/write ratio?", "are we optimizing for latency or throughput?", "is this user-facing or internal?" passes it before writing a single component.

This is not procedural box-checking. Interviewers are measuring whether you understand that requirements drive architecture, and that building the wrong thing with perfect design is still a failure.

Salary Ranges and Negotiation Context for SDE-2/MLE in India

These ranges reflect current market data from Glassdoor, levels.fyi India data, and community reports. CTC = base + variable + stock amortized over 4 years.

The stock component is the most misunderstood part of Indian product company comp. For Flipkart and Razorpay, ESOPs are illiquid until an IPO or secondary sale event — meaning your "₹50L CTC" may actually be ₹32L in real annual cash flow for 3–4 years. For Coupang, Microsoft, and Uber, RSUs vest quarterly into publicly listed shares you can sell. That liquidity difference is worth accounting for when comparing offers.

Negotiation leverage in India is highest when you have a competing offer — specifically a competing offer with a public company stock component. "Coupang has offered me X with liquid RSUs" is a stronger negotiation anchor against a Flipkart ESOP offer than any verbal claim about your market value.

How to Prep for a Specific Company Loop

The engineers who get SDE-2 offers at these companies do not prep "product company interview questions in India" generically. They prep by company, by round.

For Flipkart: 3 weeks of LLD machine coding with a real IDE. Practice cab booking, splitwise, parking lot, and food delivery system implementations — not class diagrams, actual compilable code. Then 1 week of HLD for Indian e-commerce scale problems.

For Coupang: 3 weeks of LeetCode Hard (graphs, DP with constraints, sliding window variants). 1 week of ML system design: feature stores, real-time serving, demand forecasting architecture.

For Razorpay: 2 weeks of distributed systems depth — idempotency, saga pattern, outbox pattern, exactly-once semantics. Read the Stripe and Square engineering blogs; they cover exactly what Razorpay tests. 1 week of LLD for payment flows.

For Microsoft: 2 weeks of behavioral prep with the STAR method at SDE-2 depth (specific metrics, specific conflicts, specific technical decisions). 1 week of system design for collaboration and communication tools.

Company-specific preparation is not harder than generic prep — it's more efficient. You're replacing 8 weeks of generic practice with 4 weeks of targeted practice against the actual interview.

The engineers who crack product company interviews in India in 2026 are not the ones who practiced the most. They're the ones who recognized that Flipkart's machine coding round, Coupang's ML system design round, and Razorpay's distributed systems depth probe are three completely different tests — and treated them that way from day one of their prep.