The call itself — everything except the answers

Answers — say these out loud is what to say when they ask a technical question. This is everything around it: the opening, the self-intro, the transitions, how to handle being pushed, and how to close.

You do not control which questions come. You do control these parts completely — so they should be the most fluent English of the call.

0–3    hello, they describe the role
3–5    your self-intro
5–9    last project
9–22   technical questions   -> answers.md
22–26  follow-ups, drilling
26–30  your questions, closing

0–3 minutes: the opening

Say very little. Sound relaxed. This part is not scored on content, only on whether you seem easy to talk to.

If the audio is unclear, fix it immediately. Do not spend twenty minutes guessing at half-heard questions out of politeness.

“Hi, good to meet you. Can you hear me clearly?”

“Sorry, you cut out for a second there — could you say that again?”

When they describe the role, listen for three things and write them down: what the AI part actually does, what is already built, and what is broken. You will reuse all three later.

React so they know you were listening, not just waiting:

“That’s useful, thank you. The part that stands out to me is the RAG pipeline — that’s the closest to what I’ve been doing.”

“Can I ask one thing before we start — is the AI part already in production, or is it still being built?”

Ask it early. The answer changes how you frame every later example, and it makes you sound like someone evaluating the work, not only being evaluated.

3–5 minutes: your self-intro

The one part of the call you can fully rehearse. Say it out loud ten times before the call. It should be your most fluent ninety seconds, because it sets their expectation for your English for everything after.

“Thanks. So — I’m a backend engineer with [N] years in Python, and for the last [N] years I’ve been working on applied AI on top of that.

Most of my recent work has been in two areas. The first is production backend: FastAPI services, Postgres, integrations with third-party systems. The second is LLM systems — I built a RAG pipeline on pgvector, and agents with LangGraph that call internal tools.

What I’d say is my strongest point is that I treat AI work as engineering rather than as prompting. I run evaluation sets, I trace every request, and I can tell you why a specific answer was wrong, which in my experience is what separates a demo from a system you can run.

On this role specifically — the combination of Python backend and applied AI in FinTech is exactly what I’ve been doing, so it looked like a good match. Happy to go deeper on any part of that.“

Why it works: it names two areas so they can pick one, the evaluation and tracing line targets exactly what the last candidate lacked, and it hands the conversation back. Trailing off instead leaves the interviewer unsure whether you have finished, which reads as hesitancy.

5–9 minutes: last project

The transition into it is usually theirs — “tell me about your last project”. The answer is in Answers — say these out loud.

Two things about delivery rather than content:

Signal the structure before you start. It buys you thinking time and it tells them the answer has a shape:

“Sure. Let me give you the context first, then what I actually built, and then the result.”

Stop at the result. Do not keep adding detail because they are silent. Silence usually means they are writing. Finish, then stop:

“…so that was the outcome. Happy to go deeper on any part of it.”

9–22 minutes: the technical questions

The content is in Answers — say these out loud. What belongs here is the frame around each answer, which is the same every time.

Open with the shape, not the detail.

“There are two parts to that — let me take them in order.”

“The short version is X. Let me explain why.”

Give one concrete example per answer. The client’s feedback on the last candidate was that the AI knowledge was shallow. A specific number or a specific incident is the fastest cure:

“To give you a concrete example — we had a query taking [X] seconds…”

Close every answer cleanly. This is the habit that matters most for a non-native speaker, because trailing off is what makes fluent content sound uncertain:

“So that’s how I’d approach it.”

“That’s the trade-off I’d accept, and why.”

“Does that answer what you were asking?”

The last one is for when you are unsure you understood. Not weakness — it is what a senior person does instead of answering the wrong question for ninety seconds.

22–26 minutes: follow-ups and pushing

This is where they test whether the earlier answers were real. Expect “why”, “what would you do differently”, and disagreement.

When they push on a decision, do not defend it — explain the trade-off. Confidence is showing you knew the cost, not claiming there was none:

“That’s fair. The reason we went that way was [X]. The cost was [Y], and we accepted it because [Z]. If the constraint had been different I’d have chosen the other option.”

When they disagree, agree with the part that is right first:

“I’d push back on that slightly. You’re right that [their point], but in our case [constraint], which is why [decision].”

When you do not know, never stop at “I don’t know”. Bridge:

“I haven’t used [X] specifically. I’ve done the same thing with [Y], where the trade-off was [Z] — I’d expect the same question to come up with [X].”

When you need a moment, say so instead of filling with “emmm”:

“Let me think about that out loud for a second.”

That sentence is completely normal in an English-language interview. Saying nothing for three seconds sounds thoughtful; three seconds of filler sounds nervous.

If you genuinely did not understand the question:

“Sorry, could you rephrase that? I want to make sure I answer the right thing.”

Ask this once or twice at most, and never guess instead of asking.

26–30 minutes: your questions and the close

Ask two, maybe three. The list is in Answers — say these out loud — the evaluation-and-tracing one is the strongest, because the last candidate had no answer there.

React to what they tell you rather than moving straight to the next question:

“That’s interesting — so the evaluation part would mostly be mine to build. That’s the part I’d want to start with anyway.”

Ask about the practical side too. It is a part-time role and a CTO cares:

“How would you like to split the hours — a fixed overlap with your time zone, or asynchronous with checkpoints?”

Close deliberately. Do not let the call end with “okay, thanks, bye”:

“Thanks — this was useful. Based on what you’ve described, the RAG and evaluation work is where I could contribute fastest. What are the next steps from your side?”

Asking for next steps is normal and expected. Not asking reads as a lack of interest.

The phrase bank

Learn these until they are automatic. They buy thinking time and they make the English sound deliberate rather than searching.

By moment

Starting an answer

• “There are two parts to that.”
• “The short version is X — let me explain why.”
• “Let me walk you through it.”

Buying time

• “Let me think about that out loud for a second.”
• “That’s a good question — the honest answer depends on X.”

Structuring

• “First… then… and finally…”
• “I’d start with X, because…”
• “The order matters here — I’d do X before Y.”

Giving an example

• “To give you a concrete example…”
• “The clearest case I had was…”

Trade-offs

• “The trade-off there was…”
• “What drove that decision was…”
• “The cost we accepted was…”

Disagreeing

• “I’d push back on that slightly, because…”
• “You’re right that X, but in our case…”

When it goes sideways

Not knowing

• “I haven’t used X specifically, but I’ve done the same thing with Y.”
• “That’s outside what I’ve built, but my understanding is…”

Ending an answer

• “So that’s how I’d approach it.”
• “Does that answer what you were asking?”
• “Happy to go deeper on any part of that.”

Three things about the English itself

Slow down deliberately. Fluency is heard as fewer mistakes, not more words per minute. Twenty percent slower than feels natural improves your grammar and sounds more senior.

Short sentences. Two short ones beat one long one with a clause you might lose halfway through. Full stops are free.

Do not apologise for your English. Saying “sorry, my English is not so good” tells them to start listening for mistakes. If you need a repeat, ask for one directly — that is a normal thing that native speakers also do.