What Ramp's Coding Rounds Test
- Know the four coding stages in order, from the CodeSignal OA to the AI-assisted round
- Know the problems candidates got in 2026, newest first, and which lesson works each one
- Know how Ramp scores the rounds and when it does ask LeetCode-style questions
Ramp's software engineer coding rounds look like work. You build a small program in parts, and the interviewer adds a new wrinkle once each part runs. On the live screen that program usually talks to a real service over HTTP, so the failures are real too. One Ramp engineer described the onsite coding as "very practical similar to the phone screen", and 2026 candidates say the same thing in their own words.
None of it is exotic. One September 2026 candidate said most of the hour went into the HTTP client and JSON parsing, with the algorithm the easy bit.
The coding stages, in order
1. The CodeSignal OA. Apply cold and it often arrives within minutes. For backend and generalist roles it's CodeSignal's progressive format: one problem in about four levels, around 90 minutes, scored out of 600. Each level unlocks only when the last one's tests pass. 2026 versions built an in-memory database, a task manager, a recipe manager, a file store and container operations. The link often expires in a few days, and even a perfect score doesn't promise a call. The CodeSignal OA covers the format and scoring in full.
2. The live screen. About an hour with an engineer, one practical problem in several parts. Expect very few hints. The shape that keeps coming back is a crawl over HTTP: hit a URL, read what it returns, follow it, and cope as new response types and errors show up. One September 2026 screen ran in CoderPad. Use any language with a decent HTTP client. Python is the common pick. Whether you can look things up depends on who you get: one 2026 intern was allowed Python docs, another candidate was discouraged from them, and Ramp's own intern FAQ calls its interviews "open book". Ask at the start.
3. Onsite practical rounds. The 2026 virtual onsite has one or two coding rounds alongside the AI round, a project deep dive or system design, and a hiring-manager behavioral. The coding rounds feel like a longer phone screen. One January 2026 onsite handed over a small Python repo with source and pytest files in separate folders.
4. The AI-assisted round. It first showed up as "a new format" in an October 2025 report and is in most 2026 loops. You code with an AI tool, sometimes inside an unfamiliar repo, and you have to defend every change it makes. Recruiters sometimes hint at the topic, so ask.
AI is allowed in that round only. No candidate describes it being allowed on the OA or in the standard live rounds, and Ramp publishes no AI policy for interviews. Assume it's off unless you're told otherwise. The AI question starts even earlier: recruiters now ask how you use AI day to day, sometimes as a 1 to 10 self-rating, so have a concrete answer ready.
Ramp coding interview questions in 2026
Roughly newest first, each with the latest month it was reported. Questions rotate, so learn the shapes and expect the details to change.
The URL maze (latest September 2026). The phone screen problem, by a distance, and most of its reports are from the last 12 months. You start at a URL whose JSON body points to the next paths, and you keep exploring until one returns something like a "congrats" message. Later parts bring 503s that succeed on retry, dead ends, cycles, junk bodies, and in some versions a 401 that needs a key you picked up earlier. Follow-ups include the shortest route, a concurrent crawler and testing without the live service. One September 2026 candidate skipped every 503 and never found the exit. It also turned up in onsites and a July 2026 AI engineer final round. See The URL Maze.
The core fits on one screen. A queue, a visited set and a bounded retry carry most of the first two parts.
import timefrom collections import dequeimport requestsdef fetch(url, tries=5): for attempt in range(tries): response = requests.get(url) if response.status_code != 503: # 503 is temporary; anything else is the answer return response time.sleep(0.1 * 2 ** attempt) # back off before trying again raise RuntimeError("still failing: " + url)def crawl(base_url, start_path): queue = deque([start_path]) seen = {start_path} while queue: path = queue.popleft() try: body = fetch(base_url + path).json() except ValueError: continue # not JSON: a dead end if "message" in body: return path, body["message"] for next_path in body.get("next", []): # the key name depends on the prompt if next_path not in seen: seen.add(next_path) queue.append(next_path) return NonePremium question (latest September 2026). The frontend OA most 2026 candidates describe. You score each guess letter by letter, with repeated letters as the trap, cap the attempts and show a win or a loss. Some versions give you 45 minutes. It's in Frontend Builds.
The spreadsheet with dependent cells (latest September 2026). Set and get cells, support formulas that point at other cells, and keep values right when a referenced cell changes. Variants add string concatenation and an error for circular references. In 2026 it came up in a screen and an onsite, and the fullest descriptions date from 2024 to mid 2025. See The Spreadsheet.
Premium question (latest August 2026). A list of card charges comes in, and you find the ones that repeat monthly, weekly or daily. You decide what counts as recurring, which means choosing a minimum number of hits, how much the date can drift and how much the amount can vary. A senior candidate in August 2026: "No crazy algorithms, practical questions, so get your hands dirty and you should be fine." Older versions date back to 2023.
Premium question (latest July 2026). Given flight records, find the user with the most flights, then where a user was at a given time: at an airport, in the air, or unknown. Underneath, it's grouping by user and a binary search over each user's sorted flights. Every report falls between January and July 2026, in screens and onsites.
A a Premium question (latest July 2026). Requests arrive in time order with an IP each. Accept a request if that IP has fewer than the limit of accepted requests inside the window, and rejected ones don't count. It was a March 2025 OA problem and came up again in a July 2026 technical screen. Recurring charges, a Premium question and this one share one lesson.
Reconcile records in the AI round (latest July 2026). Messy records from several sources, sometimes people and sometimes a Premium question. You normalize the fields, agree the matching rules with your interviewer, merge what matches and keep the rest apart. Then you explain what you checked by hand. A January 2026 onsite and a July 2026 AI engineer screen both used it. The AI-Assisted Coding Round walks through it.
A Google Calendar day view (2023 to 2025, latest September 2025). A frontend live screen: one column of hour markers, events that span them, and a form to create, edit and delete events. It was the classic frontend screen through 2025, and in 2026 Wordle has taken over as the build to expect. It sits with Wordle and tic-tac-toe in Frontend Builds.
Does Ramp ask LeetCode?
Mostly no, but not never. The live rounds are built around practical problems, and an April 2026 candidate wrote: "Practical questions - not strictly leetcode... Worth knowing your basic algorithms - but no need to go extraordinarily deep."
Then there are the exceptions. Two April 2026 screens, one for an intern, were plain data-structures problems. Some OAs skip the progressive build entirely: a December 2025 iOS assessment was one medium string problem in 45 minutes, and an October 2025 AI product engineer got two HackerRank problems. Keep hash maps, sorting, binary search, BFS and DFS warm. They show up inside the practical builds anyway.
How the rounds are scored
A senior candidate in September 2026 called the interviews "maybe on the easier side", and that matches most 2025 and 2026 reports. The catch is that Ramp expects every part finished, quickly and cleanly. How Ramp decides covers the bar in full.
The AI round may weigh how you direct the tool a little more. Plan to finish there too, and narrate why you accept or reject each change the tool suggests. The AI-Assisted Coding Round covers how it's judged.
Run your code after every part and fix the failure before you move on. Ask a clarifying question once, then probe the live service yourself, because the interviewer often won't explain the edge cases. When a new rule lands, change the one helper it touches instead of patching every caller.
The Ramp frontend interview
Frontend candidates build in React the whole way through. The OA is usually CodeSignal, around 45 minutes to an hour, and Wordle is the build most of them describe. Some versions pair a React build with an algorithm question, and one March 2026 OA ran 90 minutes on CoderPad. The live screen has been the calendar day view or a small game like tic-tac-toe, in 45 to 60 minutes, and a 2024 frontend candidate said speed was the only thing that counted. Frontend onsites in 2026 include the AI round too.
Some frontend postings still gate the application with a capture-the-flag puzzle: decode a URL, find a hidden link in the page's DOM, fetch a word and type it out with a small React component. It dominated 2023 to 2025. It still turns up in 2026, most recently in July, but far less often. See The Application CTF.
Practice by problem family
The lessons run in study order. Start with the OA, since it's your first gate, then the screen and onsite problems, then the frontend builds and the application CTF. The last lesson holds three older problems from 2023 to 2025 for extra reps.
The CodeSignal OA
The four-level progressive format, the 600-point score, and the banking and in-memory database builds.
More CodeSignal Builds: Files, Tasks & Recipes
More progressive OA domains: the task manager, the recipe manager and the file storage system.
Crawling the URL Maze Over HTTP
The 2026 phone screen: BFS over real endpoints, 503 retries, dead ends, keys and the concurrency follow-up.
Recurring Charges, Flight Tracking & Rate Limits
a Premium question, where a user was at time T, and a per-IP sliding-window limiter.
Spreadsheet Formulas & Dependent Cells
Cells that reference cells, updates that ripple through, concatenation and cycle errors.
The AI-Assisted Coding Round
Coding with an AI tool in a repo, and the record reconciliation task from 2026.
Frontend Builds: Wordle, Calendar & Tic-Tac-Toe
The React OA and live screen builds, with the duplicate-letter scoring that trips people up.
The Application CTF
The frontend application puzzle: decode the link, find the flag, type it out in React.
Older Practice Problems (2023-2025)
From 2023 to 2025: the email inbox, the fix-the-bugs React take-home and snake_case to camelCase.
Common questions
What coding questions does Ramp ask?
In 2026 the URL maze leads by far: crawl a chain of HTTP endpoints, retrying 503s and avoiding cycles until you find the exit. Behind it come recurring card transactions, a flight tracker and a spreadsheet with dependent cells. Frontend candidates mostly get a Premium question. The AI round has used a record reconciliation task.
Does Ramp ask LeetCode questions?
Mostly not. The live rounds are practical builds, and the OA is usually CodeSignal's progressive format. But two April 2026 screens were plain data-structures problems, and some 2025 OAs were classic DSA on CodeSignal or HackerRank. Know hash maps, sorting, binary search and graph traversal well, then spend the rest of your time on practical builds.
Can I use AI in the Ramp coding interview?
Only in the AI-assisted onsite round, where you code with an AI tool and explain each change. No candidate reports AI being allowed on the OA or in the other live rounds. Documentation lookup in those rounds depends on the interviewer, so ask before you start.
How hard is the Ramp technical interview?
The problems are medium, and some candidates call them easy. The bar is the hard part, and rejections rarely come with detailed feedback.
What language should I use?
Whatever you write fastest, as long as it has a good HTTP client for the maze. Most candidates use Python, and one 2026 onsite shipped a pytest repo. Frontend rounds are React, often with TypeScript.
How do I prepare for the Ramp coding interview?
Build the maze crawler from a blank file against a small local server that throws 503s and junk at you. Then do two Premium questions under an hour each. Take one progressive CodeSignal practice run to the last level. Rehearse a session with your AI tool where you review every diff before accepting it.
Key Takeaways
- ✓Ramp's coding stages are a CodeSignal OA, a one-hour live screen, one or two onsite builds and an AI-assisted round.
- ✓The URL maze leads 2026 by far. Recurring charges, a Premium question and the spreadsheet follow.
- ✓Problems are medium. Run your code after every part and fix each failure before the next rule lands.
- ✓AI is allowed only in the AI-assisted round. Some 2026 screens and OAs are still classic DSA.
Free Ramp interview pages
Ramp interview guideKnow what Ramp asks before you walk in.
29 coding · 3 system design · 18 behavioral questions
Get their questions first.
InterviewLegend is independent and not affiliated with or endorsed by any company it covers.