Orythara's AI interviewer, ORY, holds a real conversation with every applicant — asking the questions you configured, probing deeper when an answer is vague, and never letting on how they did. You get back a scored report that quotes the candidate's own words as evidence.
No candidate install — one browser linkLive in your ATS via REST + webhooks
Senior Backend EngineerLiveREC
ORY is speaking…CANDIDATE
Question 3 of 58:12 left
Current questionFollow-up 1 of 2
You mentioned the webhook pipeline was retry-safe — how did you make delivery idempotent?
CandidateWe keyed every delivery on an event id and stored it before dispatch, so a retry hits the same row instead of sending twice.
ORYUnderstood.
11
Integrity signals monitored during every interview
3
Market profiles, including Gulf Arabic
7
Voice and transcription models to choose from
0
Candidates auto-rejected — a human always decides
How it works
Three steps, and none of them is on a calendar.
The bottleneck in first-round screening is not judgement — it's scheduling. Orythara removes the meeting without removing the conversation.
01
Configure the role
Write the questions you want asked, set a duration cap, decide how many times a candidate may redo an answer, and add any product or project names the transcriber should expect.
02
Send one link
Create the interview from the console or your ATS and hand the candidate a URL. They open it in a browser — no download, no scheduling, no interviewer calendar.
03
Read the evidence
When they finish, Orythara scores the transcript against your competencies, quotes the answers behind each score, and pushes the result to your webhook.
The interview
A conversation, bounded by your question set.
ORY does not improvise a new interview for every candidate. It asks what you configured, then goes deeper on those same questions — so two candidates for one role are genuinely comparable.
Your questions, probed properly
A vague answer doesn't slide by. ORY asks up to two follow-ups that dig into what the candidate actually said — a concrete example, a number, a “how exactly” — while staying strictly on the configured topic. It never introduces a question you didn't write.
Turn budget per question guarantees the interview always reaches your later questions, no matter how long one answer runs.
Soft duration cap is checked only when a new question would open, so nobody is cut off mid-sentence.
The candidate can't read their score
ORY's acknowledgements are a fixed, neutral set — “Thank you.” “Understood.” A strong answer and an empty one get exactly the same response. That's enforced in code, not requested in a prompt, so praise can never leak to one candidate and not the next.
No scores, results or hiring signals are ever revealed during the interview.
“I don't know” moves the interview on without judgement — never treated as a failure or a reason to end.
Built for a nervous first-timer
ORY greets the candidate, explains how it works, and then waits until they say they're ready — it doesn't open with a question. Mid-interview they can ask it to repeat something, redo an answer they fumbled, or ask their own questions before the close.
Retries are per question and unconditional — offered on a one-word answer and the last question alike.
“Repeat that question” is free and never spends a retry.
Voice or a video avatar
Run interviews as audio-only with a natural synthesized voice, or as a live video avatar. Either way the candidate's camera stays on for the record, and the whole session is captured and stored for review.
Choose the voice and the transcription model per tenant — or per interview, to trial one.
Role glossary teaches the transcriber your product, team and project names so they don't come back phonetically mangled.
The report
Every score has a quote underneath it.
An AI rating you can't audit is worthless in a hiring decision. Orythara grounds each competency in the candidate's literal words — and the quotes are checked against the transcript, so a tidied-up paraphrase is dropped rather than shown to you as evidence.
Interview reportCOMPLETE · 5/5 ASKED
A. Rahman
Senior Backend Engineer · 14 min
78
Overall
Recommendation: Yes
Technical depth
82
Problem solving
74
Communication
88
Ownership
61
“We keyed every delivery on an event id and stored it before dispatch, so a retry hits the same row instead of sending twice.”Evidence · Technical depth · verified against transcript
Verbatim evidence, verified
Each competency carries one or two short quotes taken word-for-word from the candidate's turns. Anything the evaluator paraphrased fails the transcript check and is silently discarded — you only ever see quotes the candidate actually said.
Your competencies, your thresholds
Define the KPIs that matter for a role and weight them; the overall score becomes their weighted average and the recommendation follows the thresholds you set. Leave them unset and you get the evaluator's holistic read instead.
It admits when it doesn't know
If an interview ended early or didn't cover enough of your questions, Orythara withholds the numeric scores and the hiring recommendation, and says so plainly at the top of the report. A partial interview never masquerades as a complete assessment.
Full transcript and recording
Strengths, concerns and development areas in prose, the complete turn-by-turn transcript, and the interview recording — all retrievable through the console or the API.
Integrity
We flag. You decide. Always.
Orythara watches for the things that make a remote interview hard to trust — a second face, a second voice, a pasted answer, a long unexplained pause before a suspiciously polished response. Each signal is weighted into an integrity score out of 100 and attached to the report as evidence for a human reviewer.
No candidate is ever auto-rejected by a signal. Proctoring here is log-and-flag by design: it exists to tell a recruiter where to look, not to make the decision for them.
Behavioural
Tab switches, lost focus, fullscreen exits, paste and copy inside an answer.
Camera
No face present, multiple faces, sustained look-away, camera covered or off.
Audio & timing
A second voice in the room, and answer latency that doesn't fit the answer given.
The full signal vocabulary
Multiple faces on cameraA second voice was heardPasted text into an answerLeft the interview tabNo face detectedLooked away from the screenCamera turned off or coveredExited fullscreenUnexplained pause before a long answerWindow lost focusCopied text from the interview
CriticalWarningInformationalEach carries its own weight against the integrity score.
Markets
Built for how the Gulf actually interviews.
A candidate in Dubai reaching for an Arabic word mid-sentence is not making a mistake, and an interviewer that treats it as one is measuring the wrong thing. Orythara ships a market profile for exactly that — plus a full Gulf Arabic interview, and a neutral English profile for everywhere else.
The profile drives transcription, the interviewer's own dialect tolerance, and how fairly the answer is scored. Set it per tenant, or per candidate.
ENEnglish (neutral)The default profile for international hiring.
GCCEnglish — Gulf, code-switchingTolerates Arabic words mid-sentence without treating them as errors or as a language failure.
عالعربية — Gulf ArabicA full Arabic interview: ORY asks, listens and reports in Gulf Arabic.
Pick the voice and the transcriberThree synthesized voices and four transcription models, selectable per tenant or for a single test interview.
Platform
Multi-tenant, metered, and API-first.
Orythara is built to sit behind someone else's product as easily as it sits in front of your own recruiters — every tenant isolated, keyed, branded and separately accounted for.
Tenant isolation and branding
Each tenant gets its own roles, interviews, keys and webhook target — plus its own greeting, company description and closing statement, so ORY introduces the right employer.
REST API with an OpenAPI spec
Create an interview, poll its status, pull the report, the transcript and the recording. Bearer-keyed, versioned under /api/v1, and documented in a live spec you can authorize against.
Signed completion webhooks
Get pushed an interview.completed event with the integrity summary the moment scoring finishes, with delivery attempts recorded so you can see what we sent and when.
Usage metering and cost control
Every interview accounts for its own conversation tokens, synthesized characters, avatar minutes and stored recording — per tenant, with your own rates and markup applied.
CREATE AN INTERVIEW
curl -X POST https://your-domain.com/api/v1/interviews \
-H "Authorization: Bearer sk_live_…" \
-d '{
"jobRoleId": "role_8f2c",
"candidateName": "A. Rahman",
"locale": "en-GCC"
}'
# → { id, token, interviewUrl, status:"pending" }# Send interviewUrl to the candidate. That's it —# we run the live session, you get the webhook.
The fastest way to judge this is to take one yourself. We'll set up a role with your real questions, send you a candidate link, and walk through the report it produces.