HR Agent Manual
A resident AI that works across the whole HR function — not just hiring. It intakes roles, screens and scores candidates, runs interviews, researches public profiles, handles employee cases and onboarding, and reports on everything — while every hire/reject decision stays with a human and every action stays inside the law.
1. What the HR Agent does
The HR Agent is a full-lifecycle HR resident. On the recruiting side it turns a plain-language job description into a structured scorecard, ingests and scores candidates, runs interviews, and keeps a pipeline. Beyond hiring it also handles day-to-day HR: employee questions and cases, new-hire onboarding checklists, analytics, and a weekly summary email. It connects to your own tools — ATS, calendar, storage, web search — so it works inside your existing stack.
Chat, email, and document upload are the core ways you work with it. Phone and video interviewing are optional features you can switch on, not the heart of the product.
| Area | What the agent covers |
|---|---|
| Recruiting | Job intake, candidate ingestion, résumé parsing, fit scoring, pipeline funnel, integrity flags |
| Interviews | Async-text, voice phone screens (Telnyx), avatar video (Tavus), rubric-based evaluation |
| Sourcing | Web research on public professional information, with source links |
| HR operations | Employee cases with triage & escalation, onboarding checklists |
| Reporting | Analytics overview + automatic weekly digest email |
| Integrations | ATS import/write-back, calendar scheduling, storage, data sources |
2. Recruiting & pipeline
Job intake
What it does: You describe a role in plain language; the agent structures it into a scorecard of competencies and must-haves it can screen and interview against consistently.
How to: Open the recruiting panel on the agent's active page → Add a role → paste the job description → save. The agent returns the structured competencies.
Reads the JD, writes an interview scorecard, and hopes every interviewer uses the same yardstick.
Generates one competency-based scorecard up front so every candidate is measured the same way.
Go/backend depth, data modelling, incident judgment, and written communication, each with a screening question.Candidate ingestion & fit scoring
What it does: Add candidates manually, by résumé file upload, or by pasting résumé text. The agent parses the résumé in the background and scores fit against the role from 0–100, with an evidence-based rationale. The score is a support signal, not a verdict.
How to: Recruiting panel → Add candidate → attach a file or paste text and pick the role. Parsing and scoring run asynchronously, so large files won't time out; the score appears when it's ready.
Skims a stack of résumés, ranks on gut feel, and can't always explain why one beat another.
Scores every candidate against the same rubric and writes the evidence behind each score, so ranking is explainable.
Pipeline funnel
Candidates move through a visible funnel. The analytics panel shows how many sit at each stage:
sourced → applied → screening → interview_scheduled → interviewed → shortlisted → offer → hired → rejected → withdrawn
Integrity flags
What it does: The agent raises non-biometric integrity signals — likely AI-generated résumé text, a plagiarized take-home, or an email-domain problem — for you to review. These are text and metadata signals only.
3. Interviews (features, not core)
The agent generates interview questions from the job's competencies and can run interviews in three modes. Every mode ends in a rubric-based evaluation with an advance / hold / decline recommendation — advice, never a decision.
| Mode | How it works | Consent handling |
|---|---|---|
| Async text | Candidate answers auto-generated questions in their own time | No recording |
| Voice phone screen (Telnyx) | Agent calls or the candidate dials a shared number + agent code, or a dedicated number | Recording consent asked at the start of the call |
| Avatar video (Tavus) | Candidate joins a link and meets a realistic AI interviewer for a structured interview | Consent-gated on camera before anything is recorded |
Voice phone screens
With Telnyx configured, the agent can place or receive screening calls. In shared-number mode each agent has a unique code, so one number serves many agents — e.g. a caller dials 9085667890 then enters code 109921. You can also assign a dedicated number.
Schedules and runs first-round phone screens one at a time, taking notes by hand.
Runs consistent structured screens on demand, asks for recording consent up front, and produces a scored write-up.
Evaluation & recommendation
What it does: After an interview the agent scores each competency with a supporting evidence quote, writes an overall summary, and gives an advance / hold / decline recommendation. It may also add an integrity note if answers look AI-generated.
4. Sourcing & enrichment
What it does: With a web-search provider connected (Brave, SerpAPI, or Tavily), the agent gathers public professional information about a candidate — talks, open-source work, published writing — and attaches it with the source URLs so you can verify it.
How to: On a candidate row, click 🔎 Web research. Results appear with links.
Manually Googles a candidate and copies links into a doc.
Pulls public professional signals with citations in one click — and only public info, never scraped private data.
5. HR operations — cases & onboarding
Employee cases
What it does: Employees (or you) raise an HR case — a policy question, leave request, benefits query, or a sensitive matter. The agent triages it against your HR policies (your ASPR/handbook) and drafts a grounded answer. Sensitive cases — anything touching employee relations, harassment, or protected topics — are automatically marked needs-human and escalated rather than auto-answered.
How to: HR operations panel → Raise a case → subject + details. Resolve a case with a resolution note when it's handled.
Fields the same benefits/leave questions repeatedly and triages what needs a manager.
Answers routine policy questions instantly from your handbook and escalates anything sensitive to a person.
Onboarding checklists
What it does: When you record a hire, the agent can generate a role-appropriate onboarding checklist and track completion. Tick tasks off as they're done; the analytics panel shows onboarding progress.
6. Analytics & the weekly digest
Analytics overview
What it does: A live panel summarizing the whole function: open roles, candidates, average fit score, hires, time-to-hire, interviews and average interview score, open/escalated cases, onboarding completion — plus the pipeline-by-stage funnel.
Weekly digest email
What it does: Every Monday (08:00 UTC) the agent emails a week-in-review to your alert address: new candidates, interviews completed, hires, plus two action lists — candidates awaiting your decision and cases needing a human. You can also send it on demand.
How to: Analytics panel → Send weekly digest now to trigger one immediately; the Monday send is automatic once an alert email is set.
An HR coordinator compiles a Monday status email by hand, pulling numbers from a spreadsheet.
Assembles and sends the same digest automatically, and surfaces exactly what needs your attention this week.
HR week in review — 6 new candidates, 4 interviews completed, 1 hire, 2 open cases.
Awaiting your decision (2): Dana R. — 82% fit · Sam O. — 74% fit
Needs a human (1): "Question about parental-leave top-up policy"
7. ATS sync (Greenhouse / Ashby)
What it does: Two-way sync with your own applicant tracking system. Import pulls candidates from Greenhouse or Ashby into the pipeline; it's idempotent (deduped by ATS id and email) and can score imported candidates against a role in the background. Push writes the agent's assessment back onto the candidate in the ATS as a note — fit score, recommendation, and the human decision.
How to: Set the ATS provider + key in Settings. In the pipeline, click ↓ Import from ATS; on an imported candidate, click ↑ Push to ATS.
| Provider | Import | Write-back | Note |
|---|---|---|---|
| Greenhouse | ✅ | ✅ (candidate note) | Needs an on-behalf-of user id (ats_user_id) to write notes |
| Ashby | ✅ | ✅ (candidate note) | API key only |
Copies candidate data and interview notes between the ATS and a spreadsheet by hand.
Imports candidates once (no duplicates) and writes its scored assessment straight back onto the ATS record.
8. Interview scheduling (Cal.com)
What it does: With Cal.com connected, the agent reads open interview slots from your event type and books one for a candidate — sending the invite, attaching the meeting link and time to the interview, and emailing the candidate a confirmation. You can also share a self-serve booking link for the candidate to pick their own time.
How to: In Settings, choose Cal.com, add the API key, the event-type id, and (optionally) a public booking link. On a candidate, click 📅 Schedule and pick a slot — or send the self-serve link.
Emails back and forth to find a time, then manually creates the calendar event and invite.
Offers real open slots, books the chosen one, and emails the confirmation with the meeting link automatically.
9. Bring-your-own tools & settings
You bring your own tools of preference — the agent stores your credentials per hire and uses them on your behalf. Everything is editable after hiring, from the Settings panel on the agent's active page.
| Tool | Providers | Used for |
|---|---|---|
| ATS | Greenhouse, Ashby (Lever/Workable stored) | Import & write-back |
| Calendar | Cal.com | Interview scheduling |
| Web search | Brave, SerpAPI, Tavily | Candidate enrichment |
| Email verification | ZeroBounce, NeverBounce | Email deliverability (not "fake person" detection) |
| Avatar video | Tavus | Structured video interviews |
| Phone | Telnyx | Voice screens (shared code or dedicated number) |
| Storage | Your own S3-compatible bucket | Résumés, recordings, transcripts |
| Data source | Your DB / export / spreadsheet | Reading existing HR data |
10. Compliance & safety
Compliance is a design constraint, not a feature bolt-on. The agent is built so that the lawful, fair path is the only path.
| Principle | How it is enforced |
|---|---|
| Human-in-the-loop | The agent never sets hired/rejected. Only the decision endpoint records a human decision (with who and when). Recommendations are advice. |
| No biometrics | No facial, emotion, gaze, or lie-detection analysis anywhere — prohibited by the EU AI Act (Art. 5) and deliberately not implemented. |
| Right to erasure | GDPR/CCPA data-subject erasure removes a candidate’s data on request (soft-delete tombstone hides them everywhere). |
| Bias audit (NYC LL144) | Bias-audit export plus a candidate-requested manual (non-automated) alternative process. |
| Recording consent | Voice and avatar interviews ask for consent before anything is recorded (Illinois AIVIA, GDPR). |
| Data retention | Résumés, recordings, and transcripts are swept on a retention window (default 30 days; hired candidates exempt). |
11. Communicating with your agent
You work with the agent the same way across everything: chat, email, and document upload — voice and video are just interview features on top.
Session email address
Each hire has a dedicated email address shown on the active page (e.g. hr-a3f9b2c1@agents.boboyka.com). Email it a question, a new role, or a résumé attachment and the agent acts on it.
To: hr-a3f9b2c1@agents.boboyka.com
Subject: New role + a résumé
Please open a "Customer Success Lead" role — SaaS background, owns renewals, strong writing. Résumé attached — score this candidate against it and tell me if they're worth a screen.
In-session chat panel
The active page has a live chat panel for quick questions and requests. Chat and email share one history — anything asked by email shows up in chat and vice versa.
What you can ask the agent to do
| Request type | Example |
|---|---|
| Open a role | "Create a role for a Senior Data Engineer — Spark, dbt, owns the warehouse." |
| Screen a candidate | "Score the attached résumé against the backend role." |
| Run a screen | "Set up a phone screen for Dana and send the recording-consent notice." |
| Schedule | "Book Sam into an interview slot next week and email the invite." |
| Answer an employee | "What’s our parental-leave policy for a 2-year employee?" |
| Report | "Send me this week’s HR digest now." |