AI interview notes for hiring managers and technical interviewers

TrackHeros is an AI note taker for interviews. It joins the interview on Google Meet or Microsoft Teams as a visible participant, records and transcribes it with real speaker names, and writes a summary in an Interview profile built for the purpose: candidate, role, interviewers, an assessment split into strengths, concerns and gaps, and the recommendation in the interviewers' own words. It is for hiring managers, technical interviewers, and freelancers who run interviews on behalf of their clients.

Why are interview notes on Google Meet and Microsoft Teams so hard to take?

In an interview you are doing two jobs at once. You are listening to a person who is nervous and trying to give them a fair hearing, and you are trying to write down enough to defend a decision next week. Do the second job well and the candidate talks to the top of your head. Do the first job well and your notes say "good on systems design, ask the others" and nothing else.

Notes written afterwards drift. By the time the debrief happens two interviews later, the candidate's answer on caching has merged with someone else's, and the panel is arguing about what was said rather than about what to do. A generic meeting summary does not help much either. It will happily produce a confident paragraph about a salary expectation that was never discussed, or a recommendation the model inferred from tone.

Interview notes on Google Meet and Teams need to do a narrow thing well: capture what was actually said, attribute it to the person who said it, and leave the judgement to the people who were in the room.

How does an AI note taker for interviews work?

1. Connect your Google accounts

Connect the Google account whose calendar holds the interviews, and assign it to a workspace. If you interview for several client companies, connect the account each company gave you and assign it to that company's workspace; an account belongs to exactly one workspace, enforced in the database. Calendar sync is push-based, so a loop scheduled by a recruiter appears without you doing anything. Connecting a Microsoft account is not yet available; a Teams interview is joined from its link wherever it appears in a Google calendar invitation, including a forwarded Teams invite.

2. The bot joins

About two minutes before the start, a participant called TrackHeros Notetaker joins the Meet or Teams call. The name can be changed per workspace, and it can post a short greeting in the meeting chat when it arrives, which is a reasonable place to say that notes are being taken. It is in the participant list like any other attendee, and it waits to be admitted if the room requires it. Auto-join is decided per meeting, per recurring series, then by the account default, and the app shows which one applied, so a "do not record" interview is one switch.

3. Notes land in the right workspace

When the recording is ready the transcript is made in two passes: platform captions first, which carry the real names of the interviewers and the candidate, then an audio pass over the recording for the authoritative text. The Interview profile writes the summary from that transcript and stores it in the workspace the account belongs to. Client A's candidates never sit in Client B's list.

4. Ask later

Before the debrief, open the workspace and ask what the candidate said about the outage they handled. The answer comes with a citation to the moment in the recording, so the panel can hear the answer rather than argue about it.

What do you get after the interview?

The Interview profile produces a fixed set of fields, listed below, and each one follows the same rule: it is filled from the transcript or it is left empty. There is no field the model is allowed to guess.

Field	What is in it
Candidate	The candidate's name as heard in the interview
Role	The position being interviewed for
Interviewers	Who asked the questions, from the participant list and the captions
Assessment	Strengths, concerns and gaps, each drawn from what was said
Salary expectation	Filled in only if a figure was actually said aloud; otherwise empty
Notice period	The same rule: a stated period, or nothing
Recommendation	The interviewers' own words, never the model's opinion
Next step	The step that was agreed on the call, if one was
Transcript	Real speaker names; a line that cannot be attributed is marked Unknown rather than guessed
Minutes	A document with candidate, assessment and recommendation sections, exportable as PDF
Recording	Playable inside the app through short-lived, meeting-bound links; no public URLs

Kept separate by design

If you interview for one company, the workspace is simply where your interviews live. If you are a freelancer who runs technical interviews for several clients, it matters more: each client company's candidates stay in that company's workspace, every query is scoped to one workspace before it runs, and an unscoped query is an error rather than a leak. Postgres row-level security sits behind that as a second layer. The client isolation page explains the mechanism, and the privacy section of the homepage says what we do with your data.

What does the Interview profile actually write?

The Interview profile is one of two summary profiles in TrackHeros; the other is Project meeting. It extracts a fixed set of fields, and the rule for every one of them is the same: from the transcript, or empty.

Candidate, role and interviewers come from the invitation and what was said. Assessment is split three ways. Strengths are the things the interviewers said went well. Concerns are the things they said did not. Gaps are the things they said were missing or untested. Each of those is drawn from what was said in the interview, including the interviewers' remarks after the candidate left if they stayed on the call.

Salary expectation and notice period are the fields that generic summaries get wrong most often. Here the rule is blunt: if money never came up, the field stays empty. A number appears only if someone said a number. The same applies to the notice period.

Recommendation is the interviewers' words, not the model's. If the panel said "strong hire for the platform team, would not put them on the mobile work", that is what appears. If the panel said nothing conclusive, the field says nothing conclusive. The model does not weigh the evidence and pronounce, because a hiring decision is yours to make and yours to defend.

Next step is whatever was agreed on the call: a take-home, a second round with the CTO, a call on Friday. If nothing was agreed, it is empty.

Can I apply it to any meeting?

Yes. Every meeting in TrackHeros can be summarised as an interview or as a project meeting with one click. A workspace's minutes template sets the default, so a workspace used only for hiring can default to Interview, while a workspace that mixes engineering calls and hiring loops can be switched per meeting. Re-summarising is the same click, so an interview that was first summarised with the wrong profile is fixed by choosing the other one.

The Project meeting profile still has a place in a hiring process. The debrief itself, or the planning call where the panel agrees the role and the loop, is a project meeting: it has decisions, action items with owners and due dates, and milestones such as the offer deadline. Summarise those calls as project meetings and the interviews as interviews, all in the same workspace.

What do the minutes look like for a hiring panel?

The minutes for an interview are a document with the candidate, the assessment and the recommendation as sections, formatted in the workspace's template and exportable as PDF. Empty sections are dropped, so an interview where salary was not discussed produces a document with no salary line rather than one that invites speculation. "Copy for email" puts the same document on the clipboard as HTML and plain text, for the debrief thread. See minutes.

Who said what?

The transcript uses the real speaker names from the platform captions, so what the candidate said and what an interviewer said are never confused. Where a line cannot be attributed, it is marked Unknown rather than assigned to the most likely person. For an interview that distinction matters: a concern voiced by the hiring manager and the same words said by the candidate are two different facts, and guessing between them is worse than admitting the gap.

Correction rules per workspace handle the vocabulary that captions mangle: the internal name of the platform, the framework the role is built around, the abbreviation everyone uses. Rules apply to future interviews, and a past one is fixed by re-summarising.

The recording itself plays inside the app through short-lived links bound to that one meeting. There are no public URLs, so a link pasted into the debrief thread does not become a permanent copy of a candidate's interview that anyone holding it can open later.

Should I tell the candidate?

Yes. The bot is visible in the participant list under its own name from the moment it joins, so the candidate can see it, and the honest thing is to say what it is: a notetaker, so that you can listen instead of typing. The optional greeting in the meeting chat can say the same in writing. Recording rules vary by jurisdiction and by the candidate's employer, so treat telling them as the minimum rather than the whole answer. If the room has a lobby, the bot waits to be admitted; if a join fails, a retry button sends it again, and the app will not put a second bot into an interview that already has one. The guide to interview notes has a short script for the first minute of the call. AI interview notes on Microsoft Teams work the same way as interview notes on Google Meet: the bot joins from the link, appears in the participant list, and waits in the lobby if there is one. The Google Meet and Microsoft Teams pages cover how it appears on each platform.

Which languages does it handle?

English and Arabic, including interviews that switch between the two mid-sentence, which is common when a technical question is asked in English and answered partly in Arabic. Expected languages and the language the notes are written in are set per workspace. See the Arabic and English page.

What does it not do yet?

The items below are what hiring teams ask for first, and none of them exists today.