The AI interviewer that remembers you

Updated October 2026. The first version of this feature, described when this post launched in May 2026, ran on Cloudflare Workers and had an LLM rewrite your profile after every session. It has since been rebuilt: the profile is now computed directly from your scored answers, the product is now called Interview Prep Academy, and Pro is $15/mo. This post describes how it works today.

You did a voice mock interview on Tuesday. Zaheen — the AI interviewer — pushed back on your STAR answer because the "Result" was vague. You worked on it. You came back Friday for another session.

In most AI interview tools, Tuesday's interviewer has no idea Friday's interviewer exists. You get asked similar questions and get polite, generic feedback, and the thing you were specifically working on is forgotten.

Zaheen keeps two kinds of memory to avoid that: memory within an interview, and memory across interviews.

Memory within an interview

When you make a concrete claim — "I cut API latency from 800ms to 120ms by moving to Kafka" — Zaheen records it, with the exact words you used. A later question then builds on it:

"Earlier you mentioned the Kafka migration cut latency to 120ms — walk me through the architecture."

…instead of a textbook "What is Kafka?". Every remembered item is extracted from your own words, never inferred, so the follow-up is about what you said rather than what a model guessed you meant.

Zaheen also notices when an answer is hedged or vague ("I think maybe it was kind of…") and probes it, the way a real interviewer would ask "what exactly did you do?".

Memory across interviews

After every completed voice interview, your scored answers are folded into a rolling profile on your account:

• Per-competency averages across your last ten sessions — correctness, problem-solving approach, clarity, depth, system design and the rest of the rubric
• Trend — how each competency moved against your previous session
• Strengths and weaknesses — the competencies you're consistently strong or weak on
• Your last overall score

The profile is computed directly from your scores — no language model writes it — so the numbers mean the same thing every time and are never invented.

Two things use it:

1. Your next interview. Before the first question, Zaheen's plan receives a short note from your profile. It's deliberately a light nudge — at most one question aimed at a known weakness — so a session never turns into a re-run of your last bad day.
2. Your dashboard, which shows the profile so you can see the trend yourself. One session is a data point; seven sessions is a curve.

What this fixes

Sycophantic feedback. An interviewer with no context scores each answer in isolation. With a history, a score has a baseline: you were at 5.8 on clarity last time and 6.5 now is information; a lone "Great answer!" isn't.

Same questions, every time. Follow-ups built from your own claims, plus one nudge toward a known weakness, keep sessions from repeating each other.

Practising only what you're good at. Left alone, people drift toward comfortable topics. A profile that remembers your weak competencies applies a little external pressure.

The verbal-explanation gap. Plenty of candidates who can solve a problem struggle to talk through it. Probing hedged, vague answers is practice for exactly that.

Other things that have shipped since launch

The launch post listed two items as "not shipped yet". Where they stand now:

• Adversarial mode — shipped. A toggle on the voice interview setup screen gives you a deliberately tougher interviewer that pushes harder on vague claims and missing trade-offs.
• True interruption — partly. You can talk over Zaheen and it stops to listen, as a person would. Zaheen cutting you off mid-answer is not built.

Try it

Voice interviews with Zaheen are part of Pro ($15/mo; you can cancel any time, and payments are non-refundable — see the refund policy). The profile starts filling in after your first completed voice interview.

Not ready to pay? Timed mock interviews are free and unlimited, with no card required.

If you're a signed-in user and want to see your own raw profile, it's at /api/me/candidate-model.