Harvey Software Engineer Interview Questions

Harvey·Software Engineer·Technical Phone Screen

Jul 2026

Interviewed for a software engineering role at Harvey and got a design question that looked simple on the surface but had a pretty nasty follow-up baked in. The core problem was interesting enough that I actually enjoyed working through it, though the cycle detection part is where things got real.

Design an in-memory spreadsheet that supports setting cells to either integer values or formula expressions (referencing other cells), and implement getCell and setCell APIs. As a follow-up, detect and reject circular dependencies.

“I started with the data structure: a map from cell name to either a raw int or a list of dependencies plus an expression string.”

View Post

Harvey·Software Engineer·Onsite - System Design / Architecture

Jul 2026

System design interview at Harvey for a software engineer role, centered entirely on building a RAG pipeline for legal research. The problem was meaty and touched basically every layer of the stack, from crawling to eval. Felt like a real product they're thinking about.

Design a retrieval-augmented generation system that answers legal questions by grounding responses in publicly published memos and client alerts from large law firms, with citations back to source documents.
How do you design the crawling and ingestion pipeline to keep the corpus fresh across roughly 500 heterogeneous law firm websites without re-crawling everything constantly?
Walk through your chunking strategy and how you'd structure the retrieval index to maximize both recall and precision for legal queries.
How do you constrain the LLM to only answer from retrieved evidence, and how do you verify that citations in the generated answer actually support the claims being made?
How do you evaluate this system offline and online, and how do you isolate whether a bad answer is a retrieval failure or a generation failure?
A memo gets corrected or retracted after you've already indexed it. How do you detect this and make sure users aren't getting answers backed by stale or withdrawn guidance?
Your retrieval recall looks good in offline eval but users are still reporting wrong answers. How do you figure out where the failure is happening?
How would you extend this system to support comparative queries, like asking how different firms differ in their guidance on a specific topic?

“This is the main question and it's a beast.”

View Post

Harvey·Software Engineer·Onsite - Multi Round

Jul 2026

Harvey SWE interview, 60-minute session split between a technical deep dive on a project you own and a short behavioral block tied to company values. The project portion is where they really dig in, so if you pick something you can only describe at a surface level, you're going to have a bad time.

Walk me through a technically complex project you personally owned. What made it hard, what trade-offs did you face, and what would you change looking back?
If the core constraint that drove your main technical decision had been removed, would you have made the same choice? Walk through the alternative.
Tell me about a decision in that project you now believe was wrong. What did it cost, and how did you find out?
Describe a time you took ownership of something clearly outside your assigned scope.
Tell me about a time you disagreed with a teammate or manager. How did it get resolved, and what would that person say about the situation if I asked them directly?
Tell me about a time you failed and what you concretely changed as a result.
What is the hardest piece of feedback you have received, and what did you concretely do differently afterward?

“This is the whole first 45 minutes, so the project selection matters way more than I expected.”

View Post

Harvey·Software Engineer·Technical Phone Screen

Jul 2026

Interviewed for a software engineering role at HarveyAI and got a spreadsheet-style coding problem involving circular dependency detection. Pretty niche problem if you haven't thought about graph cycles in a while.

Implement a spreadsheet cell system where setting a cell's value can detect circular dependencies.

“The core of it is cycle detection in a directed graph, which sounds clean until you're mid-implementation and realize you need to track not just current deps but transitive ones too.”

View Post

Harvey·Software Engineer·Technical Phone Screen

Jul 2026

Harvey AI technical screen for a software engineering role, basically one big design question that spiraled into a lot of edge case discussion. Felt like a coding round dressed up as a system design question.

Design an in-memory hierarchical file system that supports adding files via absolute paths (creating intermediate directories automatically) and listing a directory's immediate children in sorted order. The system must enforce a max of 5 entries per directory, reject insertions that would exceed it, and auto-rename duplicate filenames by appending a counter before the extension. Describe your data structures, parsing logic, and time/space complexity, and walk through edge cases like root paths, trailing slashes, and invalid inputs.

“This looked like a trie problem at first and I went straight to implementing nodes with a children map.”

View Post

Harvey·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

HarveyAI's project round was one question per company value, which sounds manageable until you realize you have to actually know what their values are and have a story ready for each one.

For each of HarveyAI's company values, describe a specific experience that demonstrates how you embody that value.

“The format itself was the curveball.”

View Post

Harvey·Software Engineer·Onsite - Behavioral / Leadership

Jun 2026

Values round at Harvey for a software engineer role. They asked behavioral questions tied directly to the three values listed on their public values page, so at least you can prep for the categories in advance.

Tell me about an experience that reflects one of Harvey's core company values.

“They literally pulled the questions from the values page, so if you've read it you know what's coming.”

View Post

Harvey·Software Engineer·Technical Phone Screen

Jun 2026

Live coding round at HarveyAI for a software engineer role. The session was centered on a tree-related problem and string manipulation under real time pressure, with the interviewer pushing hard on time complexity throughout.

Solve a lowest common ancestor problem on a binary search tree, then extend it to handle string editing via simulated mouse interactions. Time complexity analysis required throughout.

“The base LCA part was fine, I know that problem.”

View Post

Harvey·Software Engineer·Technical Phone Screen

Jun 2026

Coding round at HarveyAI for a software engineer role. It was a twist on a known spreadsheet-style LeetCode problem, which I thought I had seen before until the follow-up hit.

Design a spreadsheet cell system where setCell can accept either a plain integer or a formula referencing other cells.
Follow-up: how would you detect a circular dependency between formulas, for example if A1 references B1 and C1, and C1 references back to A1?

“Recognized it as a variant of a spreadsheet problem I'd practiced before, so I felt decent going in.”

View Post

Harvey·Software Engineer·Technical Phone Screen

Jun 2026

Did a technical screen for a Software Engineer role at Harvey. The whole thing was basically one design problem: build an in-memory file system. Not the hardest thing I've done but it required more careful thought about edge cases than I expected.

Design and implement an in-memory hierarchical file system with createPath and get operations. createPath should return false if the path already exists, if the parent doesn't exist, or if the path is invalid. get should return -1 for missing paths. Also discuss your data structure choices and the time complexity of each operation.

“My first instinct was a trie and that turned out to be the right call.”

View Post