Two ways to do everything. You can build and run interviews entirely in the dashboard (no code) or over the API. Each concept page below shows both a Dashboard and an API path for creating it — pick whichever fits you. Prefer a guided walkthrough? Start with How it works, end to end.
In plain English
Think of how a real company interviews someone:- A Role is the job you’re hiring (or assessing) for — “Customer Support Specialist”, “ServiceNow Engineer”, “Account Executive”. It says what good looks like.
- A Round is one step in that job’s interview — a phone screen, a technical round, a final. A Role can have one Round or several.
- An Interviewer is the person who runs a Round — here, an AI persona with a scorecard. You build one once (e.g. an “ITOM Specialist” or a “Friendly Screener”) and reuse it in any Round of any Role.
A Role has Rounds; each Round is run by an Interviewer.You don’t need to be technical to set this up — you describe the Role in plain words, add the Rounds you want, and pick an Interviewer for each. Then you add people and the AI does the interviews.
The mental model
Configuration flows top-down. You define how the AI behaves, what it assesses, and which steps a candidate goes through — then attach people and run sessions.Reuse is the point. One Interviewer can run Rounds across many Roles; one Role can have many Rounds; one participant can run many sessions. Define your calibration once, apply it everywhere.
Interviewer
(API field name:agent_profile)
The interviewer’s personality and judgment. An Interviewer decides how the AI introduces itself, how hard it pushes, and how it converts a conversation into a score. Attach an Interviewer to a Round to control that step’s interview.
Key fields
Display name, e.g.
"Senior Engineering Interviewer".How the AI introduces itself at the start, e.g.
"Alex, a senior engineer at Acme." Sets the candidate’s first impression and the AI’s voice.Selects the recommendation vocabulary —
hiring yields hire / no-hire, training yields ready / needs practice, and so on.The scoring axes:
[{ name, description, weight }]. Weights must sum to 100. This is the rubric the AI grades against.Conversation behavior:
tone (friendly · professional · strict · challenging), style (structured · conversational · adaptive), difficulty (easy · medium · hard · adaptive), and probingDepth (low · medium · high).How scores are computed:
scale (0-10 · 0-5 · 0-100), scoringMethod (weighted_average · rule_based), and optional recommendationLogic.Grounding for the AI:
organizationContext, domainContext, scenarioContext — e.g. company background or the roleplay setup.What the scorecard contains:
includeTranscript, includeScoreBreakdown, includeRecommendation, includeImprovementFeedback (all default true).How it changes per use case
- Hiring
- Training
- Mock interview
Calibrated, slightly challenging, weighted toward role skills.
Role
(API field name:evaluation_template)
The blueprint for what you’re hiring or assessing for. A Role describes what the AI should determine and the bar for success. Every participant and session belongs to a Role.
Key fields
Display name, e.g.
"Senior Backend Engineer".The category, which drives the recommendation vocabulary and defaults.
The single thing the interview must determine, e.g. “Assess whether the candidate can own backend system design.”
The verdict label:
hire_no_hire, admit_reject, pass_fail, ready_needs_training, certified_not_certified, or custom. Auto-derived from use_case if omitted.Technical skills to probe, e.g.
["Go", "PostgreSQL", "Distributed Systems"].Behavioral traits, e.g.
["Communication", "Ownership"].What the candidate must demonstrate to pass — your bar, in plain language.
Question depth (default
intermediate).live_ai_interview, async_interview, roleplay_simulation, practice_session, or manual_review. Sets the interview format.How it changes per use case
| Field | Hiring | Training | Mock interview |
|---|---|---|---|
use_case | hiring | training | training |
objective | ”Can they do the job?" | "Are they ready to perform?" | "Are they ready for the real thing?” |
success_outcome | hire_no_hire | ready_needs_training | pass_fail |
default_session_mode | live_ai_interview | roleplay_simulation | practice_session |
skills | role-specific | competencies to build | target-role skills |
Round
(API field name:evaluation_stage)
A single step within a Role’s pipeline — e.g. Phone Screen → Technical → Final. Each Round runs its own interview with its own Interviewer, duration, and pass threshold.
Key fields
The Role this Round belongs to.
Display name shown to candidates, e.g.
"Technical Screen".ai_interview, roleplay_simulation, practice_session, manual_review, async_assessment, or final_review.The Interviewer that conducts this Round. This is where the persona and rubric plug in.
Position in the pipeline (0-based).
Minimum score required to pass this stage.
Whether a candidate can retry on failure (default
false).What happens on completion — advance automatically, or wait for a human.
How it changes per use case
- Hiring
- Training
- Mock interview
Multiple gated stages, rising bar, human approval before the final round.
Participant
The person being evaluated — a candidate, trainee, or student. Create the participant first, then schedule sessions for them. A participant is scoped to a Role, so the same email can exist as separate participants under different Roles.Key fields
Full name.
Email address — unique per account. The interview invite is sent here.
The Role this participant is being evaluated against.
Your ATS / LMS / system ID. Echoed back in every webhook payload, so you can reconcile results without storing our IDs.
{ title, organization, experienceLevel, location } — context the AI uses to tailor questions.{ skills: string[], education, languages: string[] }.Publicly accessible resume — triggers vector extraction so the AI can reference it during the interview.
Arbitrary labels for filtering, e.g.
["senior", "remote"].How it changes per use case
- Hiring
- Training
- Mock interview
A candidate sourced from your ATS, with resume and role context.
Session
The output — one AI interview for one participant, producing audio, a transcript, per-dimension scores, a recommendation, and authenticity signals. You schedule a session against a participant + template; the invite email goes out automatically. See Sessions.Putting it together
Where to next
Quickstart
Run your first AI interview in 5 minutes.
Authentication
Generate API keys and authenticate requests.
API Reference
Every endpoint with request builders and response examples.
Documentation is versioned with the platform — what you read here matches what’s deployed.