What Riwork does, and where a person decides

• The transcript. Spoken answers are converted to text; typed answers are used as typed. For a coding exercise, the final code is also read.
• Against a role-specific rubric. The skills assessed come from the role and level set for the interview and any job description, focus areas and must-ask questions the hiring team added. Each skill gets a 1 to 5 mark with quoted or paraphrased evidence, plus an overall 0 to 100 score and a recommendation.
• The output is a draft. It is written for the hiring team to read alongside the transcript.

• Facial expression, emotion, voice tone, body language or personality. Camera video is not analysed; the assessment step is also instructed not to comment on accent, voice, appearance or emotion.
• Leaving the interview tab and pasting text are counted and shown to reviewers as context. They are not scored.

A person on the hiring team reviews every candidate and makes the decision. Riwork does not reject, advance or email a candidate because of a score or recommendation. Changing a stage, rejecting, advancing and emailing a decision are actions a signed-in reviewer takes; each is written to the workspace activity log. This is checked by an automated test in the codebase (npm run test:human-review).

The hiring company is the controller of its candidates’ data and decides how it uses Riwork and what it decides about a candidate. Riwork acts as its processor. The data-processing terms are in the DPA template. [legal entity name and DPA signing route, to be completed by the owner]

• Collected: the candidate’s name and contact details as the company supplies them, answers to any screening questions, the interview transcript, coding-exercise code, activity signals (tab leaves, pastes), and a recording only if the company turned recording on and the candidate was told before starting.
• Analysed: the transcript, and code where there is one, by the AI services listed below, to write the report.
• Retained: for the workspace’s retention period, 180 days unless the company changes it. The period is shown to the candidate before they start. Expired interviews, transcripts, reports and recordings are removed by a daily job.
• Deleted on request: the company can delete a candidate at any time, which erases their interview, report and recording, and can export one candidate or the whole workspace as JSON. A candidate asks the company that invited them. The workspace owner can erase the whole workspace.
• Not stored: practice interviews store nothing on Riwork’ servers.

This is the same list, in the same wording, as on Trust & data. Always on services are part of how Riwork works for every workspace; If enabled services apply only where the company or operator turns them on.

• Anthropic

  Always on

  Runs the interview conversation and writes the assessment.

  Receives: The role details, the conversation and your name, so the interviewer can greet you — and your code, if the interview has a coding exercise.

• Deepgram

  Always on

  Live speech-to-text, when you answer by voice.

  Receives: Your microphone audio, streamed straight from your browser. By default it's opted out of Deepgram's model-improvement programme.

• ElevenLabs

  Always on

  The interviewer's voice.

  Receives: Only the interviewer's words — never your audio.

• Tavus, with Daily

  Always on

  The optional live video interviewer.

  Receives: Only the interviewer's words. Your browser joins the video call receive-only, so your camera and microphone are never sent to it.

• Vercel

  Always on

  Hosts the app, and stores recordings as private files when a company turns recording on.

  Receives: Requests to the app (with routine logs, such as IP addresses, for security and reliability) and those recordings.

• Database provider (e.g. Neon)

  Always on

  Stores workspace data.

  Receives: Employer accounts, interviews, candidates, transcripts, reports, activity signals and code from coding exercises.

• Resend

  Always on

  Sends the emails a workspace needs — invitations, reminders, password resets and decision emails — where the deployment is configured to send email.

  Receives: The recipient's name and email address, and their invitation or account link.

• Stripe

  If enabled

  Billing, only if the company subscribes to a paid plan.

  Receives: The billing contact's name and email address, and payment details. Candidate interview data is not sent to Stripe.

• Google

  If enabled

  Sign-in for hiring-team members, only if the deployment configures Google sign-in.

  Receives: Your name and email address, and a one-time sign-in code. Google returns an identity token; Riwork reads the name and email from it.

• Microsoft

  If enabled

  Sign-in for hiring-team members, only if the deployment configures Microsoft sign-in.

  Receives: Your name and email address, and a one-time sign-in code. Microsoft returns an identity token; Riwork reads the name and email from it.

• Google Calendar

  If enabled

  Live follow-up scheduling, only if a reviewer connects their own Google Calendar.

  Receives: An event on the reviewer's own calendar: the candidate's name, the job title, a time, and a link to the candidate page that needs a sign-in. No attendees, transcript or recording. Ava stores only an encrypted refresh token.

• Microsoft Graph (Outlook Calendar)

  If enabled

  Live follow-up scheduling, only if a reviewer connects their own Outlook calendar.

  Receives: An event on the reviewer's own calendar: the candidate's name, the job title, a time, and a link to the candidate page that needs a sign-in. No attendees, transcript or recording. Ava stores only an encrypted refresh token.

• Greenhouse

  If enabled

  Applicant tracking, only if the company connects its Greenhouse account.

  Receives: Candidate names, email addresses and stage names pulled from Greenhouse, and a short note with the interview score and recommendation written back as a private note. The transcript and the recording are never sent.

• Lever

  If enabled

  Applicant tracking, only if the company connects its Lever account.

  Receives: Candidate names, email addresses and stage names pulled from Lever, and a short note with the interview score and recommendation written back as a private note. The transcript and the recording are never sent.

• Workable

  If enabled

  Applicant tracking, only if the company connects its Workable account.

  Receives: Candidate names, email addresses and stage names pulled from Workable, and a short note with the interview score and recommendation written back as a comment. The transcript and the recording are never sent.

• Slack

  If enabled

  Notifications, only if the company adds a Slack incoming-webhook endpoint.

  Receives: A short message: the candidate's name, the role, a score, a stage name and a link that needs a sign-in. No transcript, no recording.

• Sentry

  If enabled

  Error reports, only if the operator sets an error-reporting destination.

  Receives: Scrubbed error text: the route, the method, the error name and message, a stack and a request id. Emails, tokens, connection strings and query strings are removed before sending, and no request body or header is included.

• Plausible

  If enabled

  Cookie-less page analytics, only if the operator turns it on.

  Receives: Page paths and event names. No cookies, no personal data as event properties.

Locations, contracts and transfer mechanisms: [to be completed by the owner]

• After an interview is complete, the candidate’s own link opens a summary of what was assessed. It contains no decision, recommendation or prediction.
• From that page the candidate can ask for something to be corrected or for a person to look again, in up to 2,000 characters. One request can be open at a time.
• A signed-in reviewer replies, marking it resolved or declined. Replying does not change the candidate’s stage, and no email is sent by the request.
• The reviewer’s reply is shown to the candidate on the same page. How quickly a company answers is the company’s own process. [response time commitment, if any, to be set by the owner]

Hiring teams can generate a draft AI-use notice from their workspace settings, covering the purpose, the role of the AI, the data used, the decision it informs, a point of contact, alternatives and retention. It is a starting draft the company edits; the product records when a notice was sent but does not decide the wording or timing the company needs. Counsel should confirm both.

It makes no claim of legal compliance, certification, accuracy, fairness or that scores predict job performance. Scores are not calibrated against a validated benchmark, and agreement with human reviewers and any group differences have not been published yet. [evaluation results and date, to be completed by owner] Companies remain responsible for their hiring decisions and for the rules that apply where they hire. This is not legal advice.