AI interview agent

Prototype, sample data

This is a Scaylar prototype built to show how an AI interview agent works, with sample data on screen. A voice agent interviews staff one at a time about the tasks that fill their week, reads back what it heard, and combines the confirmed interviews into a ranked map of where AI agents could take on work. The organization, the people and every number are made up, and the prototype uses no microphone and calls no AI service.

Open the prototype
Book a call
The opportunity map in the admin view of the interview prototype, labeled Sample data: a bar chart of addressable hours per week for eight candidate agents, from invoice intake and matching at the top to supplier payment status at the bottom, beside tiles for the total addressable hours, the number of candidate agents and the teams affected.
The prototype, with sample data. Everything on the screen is made up to show how the system works.

( USE CASE )

Who it is for and the problem

Consultancies, and the in-house teams that run change programs, when they need to know where AI agents fit across a large organization.

Workshops and surveys miss how work actually gets done. A workshop hears from a few managers; a survey gets short answers to fixed questions. The detail that decides whether an agent can help (which systems a task touches, where it gets stuck, who it is handed to) comes out in conversation, and there are never enough interviewers to talk to everyone.

The agent holds that conversation with each person, records it in a structured form and adds it to one picture of the organization.

( STEP BY STEP )

How it works

The steps describe how a production build would work. The prototype plays them back with sample data and timers.

  1. Step 01

    Interview each person

    Each person gets an invitation and talks to the agent when it suits them. The agent follows a short guide: their week, the time each task takes, where it gets stuck, the systems involved and what they would hand off. It asks follow-up questions and lets people interrupt, the way a good interviewer does. In the prototype the interview is a scripted sample played back on a timer.

    An interview in progress with an accounts payable specialist, labeled Sample data: on the left a You are speaking indicator with a waveform and the interview guide at its third topic, Where it gets stuck; on the right the live transcript of the interviewer’s questions and the participant’s answers about entering and matching invoices.

  2. Step 02

    Read it back and confirm

    At the end the agent reads back what it heard: the tasks and the hours they take, where the work gets stuck, the systems used and what the person would hand off. The participant confirms it or corrects it before anything is saved. Only confirmed summaries count toward the map.

    The read-back screen, labeled Sample data: under the heading Here is what I heard, the participant’s tasks with hours per week, where the work gets stuck, the systems they use and what they would hand off, with a Confirm summary button and a Change something button.

  3. Step 03

    Record the interview in a structured form

    In a production build, each confirmed interview becomes a record: tasks with hours per week, the systems each one uses, the hand-offs between teams, an automation fit with the reason for it, and the quote behind each point. The full transcript stays attached, so any line of the record can be checked against what was said.

    The record of one confirmed interview in the admin view, labeled Sample data: the team, date, length and status, a summary of the participant’s week, totals at a glance, the systems mentioned, and the first task with its hours per week, systems, hand-offs, automation fit and the quote it came from.

  4. Step 04

    Follow the campaign on one screen

    The admin view shows who has been interviewed and who has not, how much work has been mapped, and which teams are not covered yet. The people running the discovery can see the gaps and send reminders before the results are read.

    The campaign overview for a made-up organization, labeled Sample data: tiles for interviews completed, teams covered, hours per week mapped and candidate agents, and a table of the latest interviews with each participant’s role, team, date, length, status and hours mapped.

  5. Step 05

    Rank the candidate agents, with the evidence

    In a production build, the confirmed interviews are combined into a ranked list of candidate AI agents, each with the hours per week it could take on, the teams involved and how well the work suits an agent. Select a candidate in the prototype to see the quotes and interviews it is built from, so every line of the map can be traced back to what people said.

    The ranked table of candidate agents with addressable hours per week, teams affected, number of supporting interviews and fit, beside the evidence panel for the top candidate, labeled Sample data: what the agent would do, why it fits, and quotes from participants with the interview each came from.

( WHAT WE AIM FOR )

Targets for a production build

What a production build would be designed and measured against. They are targets, not results: the prototype runs on sample data.

  • Target: hundreds of interviews in a week, run side by side, without adding interviewers.
  • Target: under $5 per 20-minute interview in model and voice costs.
  • Target: every candidate agent backed by quotes from confirmed interviews.

( THE BUILD )

What we would build it with

A large language model (LLM)

To follow the interview guide, ask follow-up questions, write the read-back and turn each transcript into a structured record.

A voice layer

Speech-to-text and text-to-speech with interruption handling, in the browser or over the phone.

Your own systems

Invitations sent from your staff directory, and records stored where your team can query and export them.

A review step

Each participant confirms their own summary, and an analyst reviews the candidate agents and their evidence before anything is presented.

Logs and cost tracking

Every interview keeps its consent, transcript, record and cost, so the campaign’s spend and every number on the map can be checked.

( OUR WORK )

Built on work we have shipped

  • Live AI voice interviews with interruption handling and a five-part scoring rubric, for a hiring platform.
  • An AI receptionist that books onto a business’s own calendar during the call.

See our work →

Open the prototype

It runs in your browser on sample data: nothing you click or type is sent anywhere. If you have a process like this one, book a call and we will tell you plainly whether an agent is the right tool for it.

Open the prototype
Book a call

More case studies: AI trend-report agent · AI learning guide

The interview on a phone: the Scaylar prototype and Sample data labels, the participant’s role, the You are speaking indicator with a waveform, and the five topics of the interview guide with the third one current.