Guides · Interview formats and question types

Tell me about yourself: the software engineer answer

2026 · ~6 min read · OfferPilot AI

Write it three times, because three different people ask it. Forty-five seconds for the recruiter: scope, stack, why you are looking. Ninety seconds for the hiring manager: one project in real depth, aimed at their posting. Sixty seconds for the peer interviewer: what you actually build on a normal Tuesday. All three run on one skeleton, present then past then future, and the reliable way to get all three wrong at once is to start at university and work forwards.

The skeleton: present, past, future

Present, about half the clock

What you own now, stated as scope rather than as a job title. "I own the ledger service, roughly 40 million writes a day" lands because it is not on your CV. "Senior backend engineer" does not, because they read that already.

Past, only what explains the present

One or two sentences, each earning its place by making your current job look inevitable. If an old role explains nothing about the engineer in front of them, it gets no sentence. A bridge, not a history.

Future, one sentence, naming their role

Why you are in this conversation, phrased as something you want rather than something you are escaping. Almost everyone drops this beat, and it is the only one that answers what is really being asked.

At 140 words a minute, a normal speaking pace, 45 seconds is about 105 words, 60 seconds about 140, 90 seconds about 210. Split the short version roughly 25 seconds present, 10 past, 10 future. The long version adds no fourth beat. It spends 55 of its seconds inside one project.

Who is listening, and what they are scoring

RoundLengthWhat they are doingOpen withEnd onCut
Recruiter screen45 secMatching you to a checklist: years, stack, seniority, notice, bandYears, language, domainWhy you are looking, unpromptedArchitecture detail. They cannot score it.
Hiring manager90 secTesting whether you have done the work in their advertThe scope you ownThe line in their posting you are aiming atYour first job, unless it explains the project.
Peer or team fit60 secJudging whether they could pair with youYour normal week, present tenseSomething in your stack that annoys youThe pitch voice. Nobody asked for a pitch.

Peers are often handed no brief and open with this question to fill the first minute, so a polished recruiter pitch in that seat sounds like a sales call. The silent note is "rehearsed".

The same engineer, three answers

One set of raw material, three uses. Nothing is invented between versions, only shown at a different resolution.

  • Five years backend, three of them on payments. Go and Python.
  • Owns the ledger service: Postgres and Kafka, roughly 40 million writes a day.
  • Two years in data engineering at a logistics startup before that.
  • Cut the write path p99 from 900ms to 120ms by moving the idempotency check into the database and batching the consumer.
  • Leaving because the rewrite shipped and the roadmap ahead is maintenance. Target: senior backend on a high-throughput ingestion team, and the posting names latency and mentoring.

1. The recruiter, 45 seconds

Backend engineer, five years, mostly Go and Python on payments infrastructure. Right now I own the ledger service at a payments company, Postgres and Kafka behind it, around 40 million writes a day. Before that I spent two years in data engineering at a logistics startup, which is where the pipeline side comes from. The thing my team calls me for is latency work. I'm looking because the ledger rewrite has shipped and the next year and a half of that roadmap is maintenance, and your posting is a high-throughput ingestion team, which is the work I want more of.

One scope figure, no architecture, and the reason for leaving volunteered before it is asked. A recruiter relays this to the hiring manager from memory, so hand over a sentence that survives paraphrasing.

2. The hiring manager, 90 seconds

Backend engineer, five years, the last three on payments infrastructure in Go and Python. I own the ledger service at a payments company, about 40 million writes a day on Postgres and Kafka.

The piece I'd point you at, since your posting leads with ingestion latency: eighteen months ago that write path had a p99 of 900 milliseconds and it was the top complaint from two internal teams. Everyone assumed Postgres was the bottleneck. It wasn't. I instrumented the path and found about 70% of the time was our own idempotency check, a separate round trip to Redis before every write. So I moved the idempotency key into the write itself as a unique constraint, so the database enforced it in one trip, then batched the consumer into 200-row transactions. p99 went to 120 milliseconds and held through a peak week at three times normal volume.

Your description also mentions mentoring, and that's the part of the job I'd want to keep. I run our on-call onboarding and do weekly design reviews with two juniors.

What I want next is your problem specifically, ingestion at a scale where batching tricks stop paying and the design has to change.

Look at the shape of the middle beat: what everyone assumed, what was actually true, what you did, what happened. That is what goes on a scorecard. The mentoring lines exist only because the posting put them there, and the last sentence quotes the advert back without announcing it.

3. The peer interviewer, 60 seconds

Day to day I'm in Go, in a ledger service, and maybe half my week is write-path work: idempotency, retries, the ordering guarantees we promise the reconciliation team. The rest is less glamorous. Reviews, and about a day a week on call across eleven services, two of which are mine.

Stack is Go and Python, Postgres, Kafka, Kubernetes, deploys through Argo. I'm the one on my team who argues for integration tests over more mocking, mostly because every bad outage we've had has been between services rather than inside one.

The thing currently annoying me is our migration tooling. Nobody set index builds to run concurrently, so adding an index on the ledger table is a scheduled two-in-the-morning event. I've got a PR open to make that the default, which is the sort of thing I end up doing on a Friday.

That last paragraph does more work than the rest of the answer. It is unpolished, specific and impossible to fake, which is what a peer listens for, and it hands them something to ask about out of genuine curiosity.

The mistake: telling it in the order it happened

The same engineer chronologically, which is where most people land under mild pressure.

So I graduated in 2021 with a computer science degree, and my first role was at a logistics startup doing data engineering. Two years there, mostly ETL pipelines in Python, and I learned a lot about Airflow. Then I wanted to get closer to product, so I moved into backend, and I've been at a payments company since, working on a few things but mainly the ledger.

  1. The strongest fact arrives at second fifty. A service taking 40 million writes a day is the best thing about this candidate, and it lands last, phrased as "a few things".
  2. The least relevant job gets the most airtime. Chronology allocates time by age rather than relevance, so old ETL work gets three sentences and the current role half of one.
  3. It never reaches the future beat. The answer trails off in the present tense and the interviewer still does not know why this person applied.

The fix is not talking faster. Start at the present, then use the past only where it explains something. The logistics job survives in the 45-second version as eight words because it explains the pipeline instinct, and that is its entire job. Drop university unless you graduated within about two years, in which case it moves into the present beat. One sequencing note: if a project is your planned behavioural story for later in the loop, name it here in one line and no more, or you will be repeating yourself at minute forty.

Where a live copilot is the wrong tool

Worth saying plainly, given who publishes this page. This is the worst question to lean on live AI help for, OfferPilot AI included. You know it is coming, and it opens nearly every round you will sit. Drafted on the spot it is generically competent at best, and reading one off a screen gives you the flat cadence people notice in the first minute, before you have any rapport to spend on it.

The harder limit sits underneath. The future beat runs on facts that are not in your CV and cannot be inferred from it: the reorganisation, the rewrite that has just finished, the city you are moving to, the wish to stop being the only person who understands one service. No model knows any of that until you write it down yourself.

Live help earns its keep later in the same call, on what you could not predict: the follow-up about a system you touched three years ago, the design question out of nowhere. That is what Interview Copilot is built for, and it is Windows 10 and 11 only, so on a Mac the question is moot. For the opener, the tools are a voice recorder and a stopwatch. Time how long it takes to reach your first concrete fact. Over ten seconds means cut from the front.

Write the 90-second version first, even though you will deliver the 45 far more often, because the long one forces you to choose the project you are betting on. The recruiter answer is a compression of it, and the peer answer is the same material in the present tense with the polish sanded off. Rewrite only the closing sentence per company. The company Q&A generator builds a question bank for a specific company and role, and the interview questions bank covers what tends to follow the opener.

Related guides

Know what is coming after the opener.

Generate a question bank for the exact company and role, then rehearse against it.