Sixty to ninety minutes to build working, extensible code from a vague problem statement, then defend the design. The round that decides most Indian product company offers.
A machine coding round hands you a loosely specified problem - a parking lot, a splitwise clone, a snake and ladder game - and 60 to 90 minutes to produce code that actually runs. Not pseudocode, not a class diagram: a working program with a driver you can demonstrate. Then you spend 15 to 30 minutes defending the design while the interviewer proposes a new requirement and watches how much of your code has to change to absorb it.
It is the round that separates the Indian product company loop from the US big tech loop, and candidates who prepare exclusively from American material walk into it cold. Flipkart, Swiggy, Zomato, CRED, Razorpay, PhonePe and most well-known Indian product companies weight it heavily - at several it carries more weight than the DSA round. Preparation aimed at Google-style interview prep leaves it entirely untouched.
The scoring surprises people. Getting the program finished matters less than the shape of what you finished. An interviewer will take a partially complete solution with clean class boundaries over a feature-complete one written in a single 400-line file, because the whole point of the round is to see the code you would write on a team. Compiling, running, and having no obvious dead ends is the bar; elegance under a new requirement is the score.
5-10 min
The statement is deliberately incomplete. Ask what is in scope, then say out loud what you are choosing not to build. Silence here reads as not noticing.
10 min
Entities, responsibilities and interfaces before any logic. A few minutes here decides whether the last twenty are spent extending or rewriting.
45-60 min
Working code with a runnable driver. In-memory storage is expected; no database, no framework, no build tooling unless asked.
5 min
Run it. A solution that does not execute scores far below a smaller one that does.
15-30 min
A new requirement arrives. How much of the code has to change is the actual result of the round.
Baseline
Compiles, executes, handles the happy path end to end with a demonstrable driver.
The real test
The new requirement is absorbed by adding a class or an implementation, not by editing five existing ones.
Senior signal
Names that explain themselves, responsibilities that do not overlap, no clever indirection nobody asked for.
A 60 to 90 minute round where you build working, extensible code from a loosely specified problem - a parking lot, a ride-hailing matcher, a splitwise clone - and then defend the design while the interviewer adds a new requirement. Unlike DSA, the output is a running program rather than one optimal function.
They test the same thinking with different artefacts. An LLD round can be satisfied with a class diagram and a discussion; a machine coding round demands code that compiles and runs. In practice machine coding is LLD with a compiler watching, which is why the same problem set prepares you for both.
DSA asks for one function with optimal complexity. Machine coding asks for a small system with several interacting classes, where algorithmic cleverness is almost irrelevant and structure is everything. A candidate strong at DSA can score badly here, and often does the first time.
No, and this is the most common misconception. Interviewers routinely pass incomplete solutions with clean boundaries over complete ones written as a single long file. What is required is that it runs and has no obvious dead ends; what is scored is how well it absorbs the new requirement at the end.
It is standard across Indian product companies - Flipkart, Swiggy, Zomato, CRED, Razorpay, PhonePe and similar - and at several it carries more weight than the DSA round. US big tech loops usually have no equivalent, so preparation built around Google or Meta material skips it entirely.
Almost never. In-memory storage and plain language features are the expectation, partly to keep the round self-contained and partly because reaching for a framework hides the design decisions the round exists to examine. Bring a language you are fluent in and nothing else.
The LLD problem set is the same problem set machine coding rounds draw from. Build one, then have the design pulled apart.