1. The question itself
Stitched back together from the fragments it arrived in, classified — technical, behavioural, situational — and split when it is really two questions wearing one coat.
AI interview assistant
An assistant is only as good as what it knows about you.
Any model can write a decent answer about leadership. Only one that has read your CV can answer with the team you actually turned around, at the place you actually did it. This is what gets put in front of it, and why each piece is there.
Nothing is uploaded twice: your CV and documents live in your account and are reused by every session.
Context
Seven sources, assembled per question
Not a profile written once and attached to everything. Each answer is built from these, weighed against the question that was just asked.
2. The last few turns
So “and how would you test that?” is answered about the thing you were just discussing, rather than in the abstract. Interviews are conversations; a question without its conversation is a different question.
3. Your instructions
A few lines you write for the session: the one project to steer towards, the one to avoid, a title to use, a weakness to frame carefully. They outrank the defaults.
4. Your CV
Parsed into roles, dates, skills and achievements, so the answer can cite the right employer and the right year. Keep several versions and choose per session. How CV context works →
5. The job description
What they said they want, which is what the answer should be aimed at. Kept in a library, so the same role at four companies is pasted once. How JD context works →
6. Your documents
Project write-ups, architecture notes, a portfolio, last year's review — searched per question so the specifics come from your own material. How document context works →
7. Your screen
A coding problem or a diagram shared on the call: one keystroke in the Windows app sends what is on screen, and the answer comes back with the code.
Why it is trimmed
More context is not better context
Every one of these has a ceiling, and the ceilings are the reason answers arrive in seconds instead of in tens of seconds.
Speed
A model reads its prompt before it writes a word. Padding the prompt with everything you ever uploaded is paid for in the one currency an interview has none of — time.
Accuracy
Given forty passages where two are relevant, a model will dutifully use the forty. Narrowing the input is how the output stays about the question.
Cost
You are charged per answer, not per token, which only works because the input is bounded. A fixed budget is what keeps one credit meaning one credit.
Predictability
A prompt whose size depends on how much you happened to upload is a product whose speed changes week to week. The budget is the same for everyone.
Shaping the answer
Say it your way, not the model's way
An answer you would not say out loud is no use, however correct. These are set per session and changed mid-answer.
| Control | What it changes |
|---|---|
| Length | A twenty-second answer, a minute, or a full walk-through with the detail an architect would want. |
| Tone | Plain and direct, or formal. Plain is the default, because it is what sounds like a person. |
| Structure | Free-form, or STAR — situation, task, action, result — for the behavioural rounds that expect it. |
| Language | Answer in the language the interview is being held in, whatever language your CV is written in. |
| Instructions | Your own standing direction for the session, in your words. It wins over the settings above. |
| Mid-answer correction | Shorter, longer, more technical, more structure — rewritten from where you have got to, not from the beginning. |
Before the call
The same context, used to prepare
The material you assemble for a live session is the material a practice session needs, which is why it is assembled once.
Job description analysis
The required skills by category, the nice-to-haves, where your CV is thin against them, and a bank of the questions this description invites.
Mock interviews
It asks, you answer, it marks the answer and says what would make it stronger. How mock interviews work →
Question history
Every question you have actually been asked, across sessions — the most honest revision list there is.
Questions
What people ask about context
What makes an AI interview assistant different from asking a chatbot?
What it has been given. A chatbot answers the question you typed with what it learned from the internet. An assistant built for interviews answers the question it heard with your CV, the job description, your own documents and the last few turns of this conversation in front of it — so the answer names your employers, your systems and your numbers rather than generic good practice.
Does it send my whole CV with every question?
No, and that is deliberate. Your CV and the job description are trimmed to a fixed budget, and documents are searched for the passages that match the question rather than sent whole. A prompt with everything in it is slower, costs more and answers worse, because the relevant paragraph is buried in forty irrelevant ones.
How does it know which documents matter?
When your attached documents fit the budget, all of them are sent in reading order and there is no guessing. When they do not, the passages are ranked against the words of the question — BM25, the same ranking a search engine uses — and the best ones are sent. Either way the answer says which document it drew on.
Can I tell it how to answer?
Yes, in two ways. Per session: how long an answer should be, how formal, whether to use the STAR structure, which language, plus a few lines of your own instructions. And mid-answer: shorter, more detail, more technical, more structure — one press, and it rewrites around what you have already said out loud.
What happens if it does not know something?
It answers from what it has, and what it has is your material. If your CV does not mention a budget, it will not claim you owned one — it will answer from the spending you did sign off. That is the behaviour you want: an invented specific is the one thing an interviewer will follow up on.
Does it help before the interview too?
Yes. The same context drives a mock interview, where it asks the questions from a real job description and scores your answers, and a job description analysis that lists the required skills, the gaps against your CV, and the questions you are most likely to be asked.
Read next
Related pages
Give it something to work with.
Add your CV and a job description, and see what it does with the first question.
Create your account