◐ ai-teacher

🎯 Interview & Career · Chapter 6

Explaining your projects in 90 seconds

“Walk me through your current project” is the question you'll be asked in almost every round — and the one candidates prepare least. Your 90-second answer decides what the next 30 minutes are about.

Study this lesson free in the app →

The basics

The project pitch is a 60–90 second spoken summary of a project you worked on: what problem it solved, what you built, one hard decision, and the result. It's the answer to “tell me about your project” — asked in technical, manager and HR rounds alike.

Why it matters: the interviewer uses your pitch to choose what to dig into. A tool list (“Kafka, Spark, Airflow, Snowflake”) invites random questions about any of them. A pitch built around a problem, a number and a decision invites questions you prepared for.

The mental model — six beats:

  1. Problem — who needed what, and why it mattered (with a number)
  2. My role — what you owned
  3. Shape — the architecture in one sentence: source → processing → storage → consumer
  4. Hard part — one decision or trade-off, and why you chose it
  5. Result — what changed, measured
  6. Hook — an open door for the follow-up you want

The pitch is an index, not the book. Detail comes when they ask.

Words to know: pitch, hook, drill-down, numbers you must know.

Explained

Pitch A: “I worked on a data platform using Kafka, Spark, Airflow, Snowflake, dbt and Tableau. We ingested data from multiple sources, transformed it and built dashboards.”

Pitch B: “Meghalaya Cabs' finance team was reconciling driver payouts by hand — two people, three days every week, and still about ₹2 lakh of errors a month. I built the pipeline that replaced it…”

A tells them nothing and invites a random quiz on six tools. B tells them why it mattered and makes them want the next sentence.

The film trailer

A film trailer is two minutes long and doesn't show the whole film.

It shows the world in one shot, the hero and what they want, one moment of real conflict, and a hint of how it turns out — then stops, so you want to see the rest.

A trailer that lists the cast, the cameras and the shooting locations would be useless, however impressive the equipment.

Your pitch is the trailer. The deep-dive is the film. Show the problem, your role, the conflict (the hard decision) and the outcome — and leave them wanting to ask.

The six beats

Roughly 15 s / 10 s / 15 s / 25 s / 15 s / 5 s. The hard part gets the most time because it's the only beat that proves judgement.

A 90-second pitch, spoken

“At Meghalaya Cabs, finance was reconciling weekly driver payouts by hand — two analysts, three days a week, and still around ₹2 lakh a month in errors. [problem]

I owned the new payouts pipeline end to end, in a team of three. [role]

Ride and payment events land in S3, a daily dbt job builds an append-only earnings ledger in Snowflake, a weekly job computes payouts, and a reconciliation job matches them against the bank's settlement file. [shape]

The hardest call was handling late refunds. Updating last week's rows would be simpler, but finance needs to see exactly what was paid. So I made the ledger append-only and booked late changes as adjustments in the current week. [hard part]

Manual reconciliation went from three days to about an hour of review, and payout errors dropped to under ₹10,000 a month. [result]

I'm happy to go deeper on the ledger design or how we made payouts idempotent.” [hook]

Numbers you must know about your own project

VOLUME      rows/day ......... ~1M rides, ~8M events
            data/day ......... ~16 GB raw, ~3 GB Parquet
            history .......... 2 years, ~6 TB
SPEED       freshness SLA .... daily by 06:00 / 15 min for ops
            job run time ..... 22 min (was 3 h)
RELIABILITY incidents/month .. 4 → 1
            tests ............ ~200 dbt tests, 12 blocking on money
COST        warehouse ........ ~₹X lakh/month (or ‘we cut it ~30%’)
PEOPLE      team size ........ 3; I owned ledger + recon

The first drill-down is almost always “how much data?” Hesitating on your own project's volume makes the whole story sound second-hand. Round numbers are fine; not knowing is not.

How pitches go wrong

Problem, role, shape, hard part, result, hook — in 90 seconds.

The pitch is an index, not the book. Know your numbers, and only hook what you can defend.

The drill-downs to prepare

After the pitch, expect some of these — prepare a 1-minute answer for each:

‘It was already there’ is a perfectly good answer for a tool choice, followed by “but I'd keep it because…” or “I'd reconsider it because…”.

Words to know

Project pitchA 60–90 second spoken summary: problem, role, architecture, one hard decision, result.
HookA closing line inviting a follow-up on a topic you've prepared deeply.
Drill-downFollow-up questions digging into one part of your project.
Shape sentenceYour architecture as one flow: source → processing → storage → consumer.
MagnitudeA rounded number (‘~1M rows a day’) that conveys scale without needing exact figures.

In two minutes

Test yourself

1. What should the first sentence of your project pitch be about?

  1. The business problem, who had it, and a number showing why it mattered
  2. The tools used
  3. Your job title
  4. The project timeline
Show the answer

The business problem, who had it, and a number showing why it mattered — The problem gives every later detail a reason. Tools first invites a quiz; the problem invites curiosity.

2. Which beat of the pitch deserves the most time?

  1. The tool list
  2. The project start date
  3. Team size
  4. The hard decision and its trade-off
Show the answer

The hard decision and its trade-off — It's the only beat that shows judgement — and it's where interviewers like to dig.

3. The interviewer asks ‘How much data did you process?’ and you don't know. What does that signal?

  1. Nothing — volumes don't matter
  2. That you may not have been close to the work, weakening the whole story
  3. That you're humble
  4. That the project was small
Show the answer

That you may not have been close to the work, weakening the whole story — Knowing your project's magnitudes is basic ownership evidence. Round numbers are fine; not knowing isn't.

4. Your project was under an NDA with a bank client. How should you describe it?

  1. Refuse to discuss it
  2. Name the bank and its systems anyway
  3. Anonymise: ‘a large private bank’, ‘~20M transactions a day’ — keep shapes and magnitudes, drop names and secrets
  4. Invent a different project
Show the answer

Anonymise: ‘a large private bank’, ‘~20M transactions a day’ — keep shapes and magnitudes, drop names and secrets — Interviewers expect anonymisation and respect it. Refusing leaves no evidence; inventing collapses under probing.

5. Asked ‘Why did you use Airflow?’ and the honest answer is that it was already in place. Best response?

  1. Make up a detailed evaluation you never did
  2. Say you don't know
  3. Say it was already the team's standard, then say whether you'd keep it and why
  4. Criticise the team's choice
Show the answer

Say it was already the team's standard, then say whether you'd keep it and why — Honest and still shows judgement. Inherited choices are normal; your opinion on them is what's being tested.

6. Trap: does mentioning more technologies in your pitch make you look more experienced?

  1. No — it invites a quiz on each and buries the problem and your decisions
  2. Yes — breadth impresses
  3. Yes, if they're modern tools
  4. Only for senior roles
Show the answer

No — it invites a quiz on each and buries the problem and your decisions — A tool list without reasons reads as buzzwords and opens six attack surfaces. One justified decision beats six names.

There are 8 questions, 3 puzzles and 5 recall cards for this lesson in the app — the recall cards come back on later days, just before you would forget.

Learn it so it stays. Read it, test yourself, then recall it on two different days. The app schedules that for you, free.

Start this lesson →