# What Legora actually tests when it interviews engineers

Source: https://www.techinterview.org/post/3233477148/what-legora-tests-when-it-interviews-engineers/
Updated: 2026-08-03 · techinterview.org

A candidate who interviewed for a software role at Legora in May 2026 got the take-home within days of the first call: build an end-to-end chat application, a fixed stack, finish inside six hours. Nobody pretended the timebox was generous. The recruiter had already said the quiet part out loud, that the company runs fast and expects people at a desk in the office every day. The take-home is a preview of the job, not a hoop you can grind your way around.

That pace makes sense once you see where Legora sits. The Swedish legal-AI company crossed $100 million in annual recurring revenue in roughly 18 months, faster than OpenAI, Anthropic, Cursor, or Wiz hit the same mark, and a $550 million Series D led by Accel put it near a $5.6 billion valuation in early 2026. Headcount went from about 40 to more than 400 in a year, with offices in Stockholm, London, New York, Denver, Sydney, and Bangalore. It is in a direct fight with Harvey for the enterprise legal market, with law-firm clients like Cleary Gottlieb, Goodwin, and Mannheimer Swartling. When a company is scaling that hard, the interview is built to find people who ship without much scaffolding.

## The take-home is the real interview

Most of the signal comes from the build. The reported assignment is a chat application, front to back, on a stack the team names for you so they can read your code the way they read their own. Six hours is not enough time to do everything well, and that is the point. They want to see what you cut, what you keep, and whether the thing actually runs when they clone it.

Treat it like a real ticket, not a coding puzzle. Get a message from a text box to a model and back to the screen first, then make it not embarrassing. A working app with plain, readable code beats a half-finished one with a clever abstraction nobody asked for. If you reach for an ORM, a queue, or a streaming response, use it because the feature needs it, not to show range. Leave a short note on what you skipped and why. Engineers who review these have seen a hundred of them, and the ones that stand out are the ones that boot on the first try and don't make the reviewer guess.

Streaming tokens to the client is where a lot of submissions quietly fall apart. If the stack pairs a React front end with a Python or Node back end, get server-sent events or a websocket working early, because retrofitting streaming into a request-response app in the last hour is how people run out of clock. Handle the empty state, the error state, and a model call that times out. Those cases are exactly the kind of thing a lawyer using the product hits on day one, and a reviewer notices when you thought about them.

## Who you actually talk to

The live rounds are short and practical. Candidate reports describe a recruiter screen, then a call with a technical or hiring manager, then the take-home, then a session where you walk through what you built. For the more client-facing roles, sometimes labeled legal engineering, there is often a demo presentation and a conversation with the Head of Legal Engineering, and those lean harder on scenarios than on code. The questions there are about how you would handle a day of real work, how you would explain an AI feature to a client, and how you would reason through a messy practical problem. Legal knowledge is not the bar. Product judgment is.

| Round | Who runs it | What it screens for |
| --- | --- | --- |
| Recruiter screen | Talent team | Motivation, tolerance for high pace, logistics like relocation and in-office expectations |
| Manager call | Engineering or hiring manager | Past shipped work, how you talk about tradeoffs, whether you can move without hand-holding |
| Take-home build | You, solo, around six hours | End-to-end chat app on a set stack; working software, code a teammate can read, product cuts under a clock |
| Build review or final | Manager and team, sometimes Head of Legal Engineering | Walk through your choices, defend them, discuss explaining AI behavior to law-firm clients |

The exact order and count vary by role and by which office is hiring, so take the table as the shape rather than a fixed script. What stays constant across the reports is the take-home carrying most of the weight and the interviewers being blunt about the working culture.

## Why the product shapes the questions

Legora is a collaborative AI workspace for lawyers. That means retrieval over large document sets, review workflows where several people mark up the same output, and citations that have to point at the right clause or the client walks. Every one of those constraints shows up in the interview if you listen for it.

Expect to be asked how you would keep a model from inventing a case that does not exist, or how you would tell a skeptical partner at a firm what the tool can and cannot be trusted to do. A strong answer is specific: ground responses in retrieved passages, show the source next to the claim, measure accuracy with an eval set instead of vibes, and design the interface so a human review step is unavoidable rather than optional. Say what you would log so you can tell a good answer from a confident wrong one. Hand-waving about how the model is really smart now is the fastest way to lose the room.

You do not need to have shipped a legal product before. You do need to sound like someone who has thought about accuracy as a product problem, not only a model problem. The teams building this care about the unglamorous parts that keep a professional user from getting burned.

## Engineering lives in Stockholm

One structural fact changes how you should target this: the company has kept its engineering team in Stockholm and has been public about not outsourcing it, even as New York, Denver, and other offices grew for sales and customer work. If you want to write code for Legora, you are most likely aiming at Stockholm, and you should assume in-person is the default rather than a perk you can negotiate away. The recruiter raises the in-office point early, and treating it as a dealbreaker after three rounds wastes everyone's time.

The upside of a small, co-located engineering org at this stage is that a single hire moves the roadmap. The downside is there is nowhere to hide, and the interview is calibrated to that. They are hiring people who will own a surface, not fill a seat on a large team.

## Comp and what to weigh

Public salary data on Legora is thin, so the safest move is to ask directly and cross-check rather than anchor on a number someone posted online. Base pay for a Stockholm senior engineer at a well-funded startup tends to land in a market band you can sanity-check against roles at Klarna, Spotify, and Truecaller, and the current job posting or a recruiter call gives you the real figure. Levels.fyi and the Swedish market threads help for the base range, less so for equity.

Equity is where the interesting decision is. A company that tripled its valuation in a year is offering options whose strike price and current mark are worth understanding in detail. Ask for the number of options, the strike, the current preferred price, the vesting schedule, and how Swedish option taxation applies to your situation, because that last part changes the real value more than most people expect. A slightly lower base with meaningful equity at a company growing this fast can beat a bigger base elsewhere, but only if you can actually model what the grant is worth and what you would owe on it.

## How to prep without burning a weekend

Build a tiny streaming chat app before you ever get the assignment. Wire a front end to a back end to a model API with server-sent events, handle the failure cases, and time yourself. Once you have done it once on your own machine, the six-hour version stops being a scramble and becomes execution. Keep a clean starter you are allowed to reuse so you are not fighting boilerplate on the clock.

For the conversations, have two or three concrete stories ready about a time you shipped something under a real deadline and what you traded away to do it, plus one clear take on where you would and would not trust an LLM in a product a lawyer bills against. That combination, someone who ships fast and thinks hard about accuracy, is the exact person this loop is trying to find. Show up as that person and the interview mostly runs itself.