Alpha AI Recruiter
First-round interviews, run by AI. Decided by your people.
Define the role, build the interview from your own question banks, and send one invite. Candidates interview on their own schedule, every answer is scored against a rubric with written rationale, and your panel reviews before a single result is released.
67%
18 min
Scoring rationale
AI scoredTechnical depth · 82%
Described connection pooling and the failure mode it prevents, unprompted. Cited a specific incident and what was changed afterwards.
Role-specific knowledge · 47%
Could not describe the team's deployment model when asked twice. Answer stayed general where the rubric required specifics.
The problem
The first round is the most repetitive hour in hiring.
It's the same questions, asked slightly differently by whoever was free, recorded in notes nobody can compare — and it's the step that decides who ever reaches a real conversation.
Screening eats the calendar
The first round is the same conversation, thirty times over — and every one of them has to be scheduled around two busy diaries.
Every interviewer asks differently
Two candidates for one role get two different conversations, then get compared as if they'd had the same one.
Notes, not evidence
Decisions rest on half-remembered impressions scribbled mid-call. When someone asks why a candidate was rejected, there's nothing solid to point at.
How it works
From an open role to a reviewed shortlist.
Three steps. Your team's effort moves from conducting interviews to judging the ones worth judging.
Define the role and the interview
Create the job, then build the interview from sections — each with its own format, time budget, question bank and scoring weight. Reuse a template across roles, or duplicate and adjust.
The candidate interviews, any time
One email invite with an expiry window. The candidate opens it whenever suits them — no coordinator, no diary tetris. An AI interviewer asks each question aloud and adapts its follow-ups.
Your panel reviews and releases
Every answer arrives scored against the rubric with written rationale, next to the full transcript. Your panelist overrides anything they disagree with and finalises — only then does the candidate hear back.
What a reviewer sees
A score you can argue with.
Every criterion carries a number and the reasoning behind it, sitting next to the transcript that produced it. The panelist changes what they disagree with, and the release control is theirs.
- Rubric scores with written rationale, per criterion
- Full turn-by-turn transcript alongside the evaluation
- Override any score — the comment is kept with it
67%
18 min
Scoring rationale
AI scoredTechnical depth · 82%
Described connection pooling and the failure mode it prevents, unprompted. Cited a specific incident and what was changed afterwards.
Role-specific knowledge · 47%
Could not describe the team's deployment model when asked twice. Answer stayed general where the rubric required specifics.
The interview itself
Four ways to answer — one transcript.
A section can be text, voice, video or live code. Whatever the format, the answer lands in the same transcript and is scored against the same rubric.
Text
A chat-style exchange. Questions stream in as they're composed, so it reads like a conversation rather than a form.
Voice
The candidate answers hands-free. Speech is transcribed live into the same transcript as everything else.
Video
Webcam on, with presence and attention signals gathered quietly in the background — never as an automatic verdict.
Live coding
A full VS Code-grade editor with language selection, a Run button that doesn't count, and a Submit that does.
What you get
A screening round that leaves evidence behind.
Rubric scores, with the reasoning
Every answer is scored against expected signals, and the score comes with the rationale behind it. At the end: an overall score, a section breakdown, and a clear recommendation.
A transcript of everything
Every question and answer is stored turn by turn with timestamps. When a hiring decision is questioned six months later, the evidence is still there.
Question banks that compound
Build banks by topic, difficulty and format. Import in bulk from CSV or JSON, or have AI draft a starting set from a topic — then edit it like anything else.
Integrity signals, not accusations
Presence, focus and paste events are gathered during a session and rolled into an advisory anomaly score. It informs your reviewer. It never rejects anyone.
Pipeline you can actually see
A live dashboard of open roles, interviews in progress, submissions awaiting review and average score — plus funnel and score-distribution reports.
Auditable by design
An append-only log records who did what, to which record, with a before-and-after of every change — actor, timestamp, IP and all.
AI you can defend in an HR review
A person signs off on every result.
Hiring decisions carry consequences, so the AI never gets the last word. It asks, it scores, it explains its reasoning — then a named human on your panel reviews the evidence, changes what they disagree with, and takes responsibility for the outcome.
Nothing releases itself
A candidate sees a result only after a panelist finalises it. There is no automatic pass or reject path.
Override anything
Scores, section marks and the recommendation are all editable, with the reviewer's comment kept alongside.
Your questions, your rubric
Interviews are built from question banks you own and signals you define — not a generic model's idea of a good hire.
Scoring can stay in-house
Run evaluation entirely offline if candidate data must not leave your infrastructure, or connect a hosted AI provider by choice.
Four consoles
Everyone gets the view they need — and nothing else.
Access is scoped by role and by department, so a recruiter in one team never sees another team's candidates.
Recruiter
Creates roles, builds interview templates, invites candidates, tracks the pipeline and pulls reports.
Panelist
Works a review queue of completed interviews — reads the transcript, overrides AI scores, and finalises.
Candidate
Accepts an invite, takes the interview when it suits them, and sees a clear result once it's released.
Admin
Manages users, roles and permissions, owns the question banks, and reads the full audit log.
Security & trust
Specific, not vague.
Every line below is how the system is actually built, not a promise. What isn't finalised — candidate notice wording, accommodations, a formal appeal process — is left off rather than assumed.
Passwords never stored in the clear
Hashed with BCrypt at cost factor 12 — a deliberately slow hash, chosen so a stolen database can't be brute-forced quickly.
Sessions rotate on every refresh
Short-lived access tokens plus a refresh token that's replaced — and the old one invalidated — every time it's used.
Every query is scoped to your organisation
Tenant isolation is enforced at the data layer, not just in the UI — one company's records are not reachable from another's session.
An audit log nothing can edit
Append-only: who did what, to which record, with a before-and-after snapshot, timestamp and IP — not just a summary line.
Proctoring signals, not verdicts
Where a recruiter enables video proctoring, signals like tab focus or multiple faces are surfaced to reviewers as advisory information — never as an automatic reject.
Media retention you control
Recordings and transcripts carry a retention date your organisation sets, for automatic cleanup after evaluation.
Who it's for
Built for teams who have to justify the decision.
In-house talent teams
Running high-volume roles where the first round is the bottleneck and consistency between interviewers is a real problem.
Departments with a panel process
Where a subject-matter expert must sign off on every hire, and that sign-off needs to rest on evidence rather than a recollection.
Regulated and public-sector hiring
Where every decision has to be defensible after the fact: same questions, same rubric, full audit trail, human accountable at the end.
Already using Alpha for assessment in a school or centre? AI Recruiter runs on the same principle — AI does the volume work, a qualified human keeps the final say.
Questions
Answers before you ask.
No. The AI conducts the interview and scores it, but a human panelist reviews every evaluation and can override any score. Nothing is released to a candidate until that person finalises it — the same human-in-the-loop principle behind our exam product.
That's your choice. The platform can run its scoring entirely offline, with no candidate answer ever leaving your infrastructure, or connect to a hosted AI provider if you prefer. Data is isolated per tenant, media has configurable retention, and every action is written to an append-only audit log.
The text interview loop works end to end: invite, interview, transcript, review, release. Voice, video and live-coding are built into the candidate experience and are being wired through to full scoring — we'll show you exactly where each one stands in a demo rather than overstating it here.
No, and it isn't trying to be. It sits between sourcing and your final rounds, replacing the first screening interview. It's designed to sit alongside whatever ATS you already run.
They interview whenever they want, without waiting a week for a slot, and they get a structured result rather than silence. The interview itself is designed to feel like a conversation — questions are spoken aloud and follow up on what was just said.
It's in active development and we're taking on early partners now. Book a demo and we'll walk through what's live today, what's landing next, and whether your hiring process is a fit.
See AI Recruiter on one of your open roles.
Book a 20-minute demo and we'll run the whole loop — build the interview, take it as a candidate, and review it as a panelist.