Harvey interview prep

See what Harvey asks, then get a plan for it

Real Harvey frontend questions and candidate reports, turned into a week-by-week prep plan.

~2 min · free preview

Company prep

Harvey Front End Interview Guide

The Harvey frontend interview process, real questions, and candidate reports.

Questions
153
Reports
Solved
—

Verified by FrontendLead

Practice questions153

From the community

Interview reports

Harvey frontend interviews almost always circle one product shape: an editable spreadsheet grid with formulas that reference other cells. Harvey.ai builds legal AI products, and the screen grades whether you can ship a dense, stateful UI under a hard timer.

This guide is built from 6 Harvey interview reports that FrontendLead members posted between July 2025 and October 2026, plus the 6 coding questions members have tagged with Harvey. You’ll get the stages, what each round grades, and a four-week plan that targets them. The exact prompts candidates were asked are in the members-only sections below.

The single biggest pattern: treat the phone screen as a race to a working grid, then a formula evaluator. Candidates who spent too long on styling or clarifying questions ran out of time before the formula follow-up.

By Shivam Bhalla · Updated

01

Harvey frontend interview at a glance

Aspect
What to expect
Timeline
Often a few weeks; confirm with your recruiter
Stages
Recruiter or hiring manager screen, then a technical phone screen
Signature topic
Editable spreadsheet / grid with cell formulas
Coding style
Practical React or vanilla JS UI; minimal styling expected
Company
Harvey.ai – legal AI platform (not a generic “Harvey”)
Languages
JavaScript or TypeScript; React is common for the grid UI
Levels seen
Frontend, Senior, Staff, and full-stack screens share the same sheet theme
Bar
Finish the base grid fast; formula support decides the round
02

The Harvey front end interview process

Harvey does not publish a detailed public frontend interview guide, so treat the stages below as the path members report. Confirm round count and editor setup with your recruiter.

  1. 1

    Recruiter or hiring manager screen

    30 to 45 minutes

    Background, why Harvey, ownership stories, and tough feedback. Public reports outside our set also describe project ownership and curiosity outside your immediate scope. Have one legal-tech or AI-product reason ready, not a generic “I like startups” answer.

  2. 2

    Technical phone screen

    About 60 minutes

    Live coding focused on a spreadsheet-like grid. Expect a two-part structure: editable cells first, then formulas that read other cells. Interviewers often de-emphasize styling and semantics so you spend time on state and evaluation. Several members said the pace felt “implement ASAP” with little room to chat.

  3. 3

    Later rounds

    Varies by role

    Member reports in this set concentrate on the screen. Full loops at AI companies often add design, code review, and behavioral rounds; confirm with your recruiter rather than assuming a fixed onsite shape.

  4. 4

    Decision

    Days to about a week

    One public write-up noted failing when later formula parts were unfinished. Budget time for the hardest follow-up, not a perfect first paint.

03

Why Harvey keeps asking for a spreadsheet

Harvey ships dense document and knowledge workflows for legal teams. A grid with interdependent cells is a clean proxy for the skills that matter: controlled inputs, derived state, dependency graphs, and talking about production hardening after the happy path.

Interviewers want to see:

  • A clear cell address model (column letter + row number)
  • Edit UX that does not fight the user (focus, commit on blur or Enter)
  • Formula parsing that pulls referenced cells and computes a result
  • A short production-grade punch list at the end: cycles, invalid refs, recompute scope, accessibility

If your prep is only LeetCode arrays, you are practicing the wrong muscle for this screen.

04

What “good enough” looks like in 60 minutes

Members who passed or nearly passed share a timing pattern. Aim for a usable grid in the first third of the slot, a working formula path in the second third, and a spoken production backlog in the last third.

  • Minutes 0 to 15: cell addresses, controlled inputs, commit on blur or Enter, render a small fixed grid
  • Minutes 15 to 40: detect = formulas, resolve referenced cells, show computed values without losing raw text
  • Minutes 40 to 60: walk through cycles, invalid references, recompute scope, keyboard UX, and how you would test it

Do not invent a full Excel clone. Interviewers grade clarity of the model under time pressure. A small correct CONCAT or SUM path beats half-finished operator precedence with no UI.

Full-stack variants may push evaluation into setCell time instead of lazy getCell reads. Ask which they want in one sentence, then commit. Switching mid-implementation costs the round.

05

What the coding rounds actually test

These are the patterns that come up again and again in Harvey reports. Each card links to the closest verified practice question on FrontendLead.

Editable N×N grid of cells

Part 1 of almost every screen

Build a 10×10 (or similar) sheet of inputs. Cells are addressable like A1, B2. Styling is optional; getting focus and value updates right matters more. One Staff candidate was told there was no emphasis on semantic elements.

Practice: Reactive Sum Sheet and Minimal Recompute Sheet.

Single formula: CONCAT of two cells

Common part 2

Support something like =CONCAT(A1,B2). A util may be provided to extract cell refs from the formula string. Keep display vs raw value straight so editing a formula does not lose the expression.

Practice: Formula Spreadsheet.

Richer formulas and dependency evaluation

Full-stack and tougher screens

Variants include =SUM(A1,B1) and multi-cell arithmetic. Members recommend DFS (or a dependency walk) for evaluation. Minimal Recompute Sheet is the practice target several candidates named explicitly.

Practice: Minimal Recompute Sheet and Reactive Sum Sheet.

Production-grade follow-up

End of screen

After the demo works, talk about cycles, batch recompute, virtualization for large sheets, undo, and collaboration. You do not need to implement all of it; show you know what breaks in production.

Practice: Design a Spreadsheet.

Async and utility fluency

Tagged practice themes

Harvey-tagged FrontendLead questions also include concurrency-limited maps and file navigation UIs. Keep async control flow warm even if your screen stays on the sheet.

Practice: Async Map with Limit, Settled Pool Map, and File Navigation.

06

Real Harvey interview questions reported by candidates

Every question below comes from a FrontendLead member’s interview report. Dates are when the report was posted.

Round Question Level
Technical screen Google Spreadsheet-like grid with edit and formulas such as =SUM(A1,B1) Full stack, Senior (Oct 2026)
Phone screen (2 parts) 10×10 editable sheet with A1-style names; then =CONCAT(A1,B2) with a util to extract cells Staff Frontend (Aug 2026)
Tech screen Build a spreadsheet with editable cells and formulas Software Engineer (Jul 2026)
Phone screen Excel-like grid Frontend (Nov 2025)
Technical phone screen Part 1 editable spreadsheet UI; part 2 formulas referencing other cells Frontend E5/E6 (Jul 2025)
Screening 10×10 inputs, then support =CONCAT(A1, B2) Senior Frontend (Oct 2025)

Members can read the full write-ups in the Harvey interview reports.

07

What Harvey system design interviewers dig into

Member reports in this set are light on formal design rounds, but the sheet prompt itself is a design conversation once formulas appear. Prepare as if you will whiteboard the client model.

Cell model and addressing

Core

Map addresses to raw input vs computed display. Decide whether formulas live as strings that start with =.

Dependency graph

Asked in tougher variants

When A1 changes, which cells must recompute? Eager vs lazy evaluation, and how you would detect cycles.

Practice: Design a Spreadsheet.

Large-sheet performance

Production talk

Virtualization, windowing, and avoiding full-sheet re-renders on every keystroke.

Practice: Design a Virtualized Grid.

Collaboration and conflict

Senior signal

If two users edit related cells, how do you show stale computed values and resolve races?

08

Worked example: formula cells without a full spreadsheet engine

Practice until you can sketch this model in about fifteen minutes. Interviewers care more about the separation than about perfect Excel parity.

// raw[addr] is what the user typed; getValue resolves formulas.
const raw = new Map(); // 'A1' -> '10' | '=CONCAT(A1,B2)'
function parseAddr(token) {
  const m = /^([A-Z]+)(\d+)$/.exec(token.trim());
  if (!m) return null;
  return m[0];
}
function getValue(addr, visiting = new Set()) {
  if (visiting.has(addr)) throw new Error('cycle');
  visiting.add(addr);
  const text = raw.get(addr) ?? '';
  if (!text.startsWith('=')) {
    visiting.delete(addr);
    return text;
  }
  // Toy CONCAT only - extend with SUM / arithmetic as needed.
  const inner = text.slice(1);
  const concat = /^CONCAT\(([^,]+),([^)]+)\)$/i.exec(inner);
  if (concat) {
    const a = getValue(parseAddr(concat[1]), visiting);
    const b = getValue(parseAddr(concat[2]), visiting);
    visiting.delete(addr);
    return String(a) + String(b);
  }
  visiting.delete(addr);
  return text;
}

Three details separate strong answers:

  • Store raw, display computed. Editing should show the formula string; the grid face shows the evaluated value.
  • Guard cycles. Even if part 2 is only CONCAT, mention cycle detection; harder variants require it.
  • Narrate recompute. Say whether you recompute the whole sheet or walk dependents after setCell.

“No emphasis on styling, so it can be as minimal as you need.” – Staff Frontend candidate on a Harvey phone screen

For the full design walkthrough, see Design a Spreadsheet.

09

Behavioral and “why Harvey”

Harvey is a legal AI company. Interviewers expect a specific reason you care about that domain: document workflows, trust and citations, or shipping AI tools lawyers will actually use. Pair that with two project stories you can defend end to end.

Strong motivation answers name a concrete workflow (contract review, research memos, knowledge retrieval) and a product principle you care about, such as verifiability or latency when a lawyer is mid-call with a client. Weak answers stop at “AI is exciting.”

For project deep dives, pick work where you owned UI state complexity: interdependent forms, collaborative editors, or data-heavy grids. Map that story to the same muscles the spreadsheet screen tests.

Use FrontendLead’s behavioral questions for STAR practice on ownership, tough feedback, and discovering problems outside your lane.

10

How the loop changes by level

Mid / E5-style

Working grid + one formula

Get editable cells and a single formula path working. Clear narration beats clever abstractions.

Senior

Speed and extensibility

Same theme, less forgiveness on time. Structure part 1 so part 2 is a small addition, not a rewrite.

Staff

Production talk after the demo

Expect a sharper “what would you improve for production” segment: cycles, recompute, collaboration, a11y.

11

A four-week Harvey prep plan

  1. 1

    Week 1: Grid UI under a timer

    Coding

    Build a bare 10×10 editable grid twice in 25 minutes. Practice: Reactive Sum Sheet.

  2. 2

    Week 2: Formulas and recompute

    Coding

    Add CONCAT and SUM-style formulas. Practice: Minimal Recompute Sheet and Formula Spreadsheet.

  3. 3

    Week 3: Spreadsheet system design

    Design

    Whiteboard a sheet for 45 minutes: model, deps, virtualization. Use Design a Spreadsheet and the frontend system design framework.

  4. 4

    Week 4: Mocks and stories

    Loop rehearsal

    Two timed sheet mocks. Write “why Harvey.ai” and two deep dives. Review behavioral questions and Harvey reports.

12

Mistakes that sink Harvey candidates

  • Polishing CSS first. Members said styling was de-emphasized; formulas were not.
  • Over-clarifying. One senior candidate felt there was almost no room to ask questions; ship a baseline fast.
  • Storing only displayed values. You need the raw formula string to round-trip edits.
  • Ignoring cycles and invalid refs in the wrap-up. Even unfinished implementations can pass the talk track if you name the hard parts.
  • Generic motivation. Harvey is legal AI; say why that product, not “AI is the future.”
13

Harvey front end interview FAQ

How long is the Harvey frontend interview process?

Member reports focus on the technical screen after an initial call. End-to-end timing varies; ask your recruiter for the full loop for your role.

Does Harvey ask LeetCode questions for frontend roles?

Reported frontend screens center on a practical spreadsheet UI, not classic algorithm drills. Full-stack variants still stay on the sheet theme in our set.

Can I use React in the Harvey coding interview?

Yes. Several candidates built the grid in React. Vanilla inputs also work if your state model is clear.

What is the most common Harvey frontend coding topic?

An editable spreadsheet-style grid with formulas that reference other cells. Exact prompts are in the members-only sections above.

Is Harvey the same as other companies named Harvey?

This guide is about Harvey.ai, the legal AI company. Confirm the company domain on your recruiter email.

Where can I practice for Harvey?