Intuit Software Engineer Interview Questions

“I would layer it. The strongest signal is what this business did before: if the owner put a merchant under supplies last month, do the same now. Those personal rules come first. When there is no history, a model trained on general patterns suggests a category from the merchant and the amount. If the system is not sure, it should ask and not guess silently, since a wrong category affects tax later. Every correction is stored as a rule for that business and also feeds the general model after review. I would measure how often owners accept the suggestion without changing it.”

“I would begin at the edges, because that is where rule code breaks. For each income level where the rate changes, I test the value on the line, one unit below and one unit above. Then I add the official worked examples as fixed cases, since they are the real definition of correct. I also like property tests: with everything else equal, more income should never give less tax, and a deduction should never raise it. Last, I keep a large file of saved inputs and outputs. If a change alters any of them, the build fails and a person must confirm it was meant.”

“At the scheduled time a job splits the list into batches and puts them on a queue. Workers take a batch, build each message and send it. The unsubscribe check happens at the moment of sending, because someone may have opted out after the campaign was scheduled. Receiving mail servers limit how fast they accept mail, so I would control the rate for each domain and retry temporary failures with delay. Each contact and campaign pair is recorded, so a retried batch does not send the same email twice. Bounces and unsubscribes flow back to the list, so the next campaign is cleaner.”

“I would first find out how often the bureau updates and what each request costs, because that sets the refresh plan. The score is stored with the date it was taken, and the app reads from our store, not from the bureau on every visit. A scheduled job refreshes each member's score at a sensible interval and spreads the load through the day. On screen I would always show when the score was last updated, so the member is not misled. The old value stays if a refresh fails. Since this is sensitive data, it is encrypted at rest and access is logged.”

“At your company we had to support a new reporting rule by a date set by the regulator. There was more work than time. I listed every item and marked what the rule strictly required and what was only comfort for users. We built the required parts first and kept the full test plan for them, because a wrong report would be worse than a plain one. A nicer summary screen was moved to the next release, and I told the product manager early so nobody was surprised. We shipped on time. The result was your result.”

“A support colleague at your company sent me a ticket from a customer whose monthly report showed a total that was off by a small amount. It would have been easy to call it a rounding matter. I asked for the customer's data, with permission, and reproduced it. The cause was a transaction dated on the last day of the month in the customer's time zone, which our server counted in the next month. Our tests all used one time zone. I fixed the date handling and added tests for several zones. Then I searched for other customers affected. The result was your result.”

“First I would stop it spreading, with a fix or by switching the feature off. Then I would work out exactly who was affected, by running the corrected calculation over saved returns and comparing. This is not something for engineering to handle alone, so I would bring in my manager, legal and support straight away with the facts I have. Customers who already filed need to be told plainly and helped to file a correction, and we should make that as easy for them as we can. Afterwards I would run a review of why our tests missed the case and add it to the test set.”

“I like engineering that takes a task people dread and makes it small. Intuit does that across several products, from tax filing with TurboTax to books in QuickBooks and campaigns in Mailchimp. The customers are individuals and small business owners who cannot afford mistakes and do not have experts on staff. So the software has to be both correct and kind. In my work on your project I saw how much one confusing total can cost a user in worry. I want to build systems where that does not happen, and to learn from engineers who have done it at scale.”