This article describes how AI features in candidate.fyi are structured from a product and data-flow perspective. It is written for administrators, privacy reviewers, and compliance teams who need a clearer picture of when AI runs, what information is involved, and where people stay in control.

Important: AI features support scheduling, coordination, note-taking, and reporting. They are not designed to score, rank, or automatically select candidates. Your organization remains responsible for employment decisions.

How AI processing fits together

Most AI features follow the same high-level pattern:

1. Trigger. Someone uses candidate.fyi (or an automated rule runs on their behalf), such as asking a question in Insights, refreshing a scheduling summary, or entering availability in plain language.

2. Context gathering. candidate.fyi collects the information needed for that task from your organization’s data—for example, interview metrics, emails and notes on a scheduling request, or the text you typed in an availability field.

3. AI request. That context is sent to a third-party AI service to generate a response, summary, classification, or structured suggestion. Processing happens to produce that specific result (inference), not as a separate step you configure in the product.

4. Result in the product. The answer or suggestion appears in candidate.fyi (a chat reply, summary card, proposed time slots, interview notes, and so on).

5. Human oversight. Depending on the feature, a person must review before anything changes in your systems, or the feature is read-only and never updates records on its own.

Organization controls (what must be turned on)

Most AI features are optional. candidate.fyi turns them on at the organization level. If a feature is off for your organization, related menu items, toggles, and cards usually do not appear.

Typical admin paths:

• Settings > fyi > fyi Agents — Interview Companion, Scheduling Agent (follow-ups / nudges)

• Settings > fyi > fyi Settings > Summarization — AI scheduling request summary

• Insights — Ask FYI and AI Query (when enabled for your organization)

• Per scheduling request — Scheduling Followup Agent under request settings

• Per interview — Interview Companion assignment and recording consent flows

AI features — architecture at a glance

Feature details — data flow and boundaries

Insights Ask / Ask FYI

Product entry points:

Insights > Ask (Ask vs Build modes; placeholder such as “Ask fyi anything about your operations.”)

Data boundary:

Questions and scoped operational data leave candidate.fyi only to generate the response shown to the internal user. The feature does not write back to your ATS or change interviews.

Retention / notice:

No separate in-product consent flow beyond the AI Query accuracy disclaimer.

Interview Companion and AI notetaker

Product entry points:

Configure under Settings > fyi > fyi Agents > Interview Companion (capture mode, consent model, retention). Assign per interview, request, booking link, or Profile > Interview Companion defaults. Candidates see Recording & AI notes with consent choices.

Data boundary:

Interview media and/or transcripts are processed to create fyi Generated content on the interview (summary, transcript, Q&A, timing insights). Ask fyi on an interview grounds answers in that interview’s recording/transcript. Optional feedback autofill suggests form content interviewers can edit.

Meeting join:

A bot may join the interview session (including dial-in flows) to capture audio/video or transcript-only content, depending on your settings—not a feature the candidate configures separately beyond consent.

Recording, consent, and retention

Administrators control how capture and consent work under Interview Companion settings:

• Capture mode: Full capture (audio/video + transcript) or Transcript only

• Candidate consent model: Opt-out (recording on by default) or Opt-in required

• Custom consent message and optional recording consent link token

• Retention windows for media/transcripts and for generated notes

Retention:

Administrators set Media & transcript retention and Notes retention on the Interview Companion settings page (options from 7 days through indefinite).

Human oversight:

In-product copy directs users to review evidence; analysis supports but does not replace interviewer judgment.

MCP connector

Product entry points:

User connects from an external MCP client, signs into candidate.fyi if needed, picks an organization on the consent screen, and clicks Allow access

Data boundary:

candidate.fyi exposes authorized scheduling and (when enabled) analytics operations to the external assistant. Any further sending of that data to that assistant’s AI models is outside candidate.fyi’s UI and policies.

AI Availability input

Product entry points:

Candidate Availability modal — Ask AI field and Generate button; Chrome extension with helper text to describe or paste availability.

Data boundary:

Only the text the scheduler enters is sent to propose slots. Candidates using the portal continue to use structured date/time pickers, not this AI field.

Human oversight:

Proposed slots are never saved until a user confirms.

Scheduling request AI summaries

Product entry points:

fyi AI Summary on the scheduling request Interviews tab; enable under Settings > fyi > fyi Settings > Summarization > AI scheduling request summary.

Data boundary:

Summaries condense request-linked notes, email content, and related activity into a single card. Refresh re-runs summarization on current context.

Scheduling follow-up assistant (Nudge)

Product entry points:

Settings > fyi > fyi Agents > Scheduling Agent (Scheduler Reminders, Candidate Followup Nudges, Interviewer Attendance Nudges, Candidate Interview Reminders, auto-close options).

Architecture split:

• Outbound email path (no generative AI): Follow-ups use predefined templates with dynamic fields (candidate name, links, etc.). AI does not draft these emails.

• Inbound email path (AI classification): When candidates reply, AI may classify intent—for example withdrawal, reschedule, or “needs a human”—to drive auto-close or routing.

Notice:

Candidate-facing messages identify an AI Assistant supporting [recruiter first name] in the signature.

Employment decisions and bias

AI features are operational aids. Interview Companion may highlight skills, communication, role fit, and risks to help interviewers review evidence, but the product explicitly keeps human judgment primary. candidate.fyi does not present AI features as automated hiring decisions, ranking, or selection.