In-house recruiters running volume
One role, two hundred applicants, and a first round that realistically reaches thirty of them. Everyone gets the interview; you get a ranked shortlist instead of a resume pile.
Kestrel
AI first-round interviewer
Every candidate gets a real interview.Every decision is made by a human.
Kestrel runs the whole first round — application, resume verification, a 30-minute two-way interview, and a ranked shortlist your hiring manager can read in five minutes.
Built for in-house recruiters, HR teams without a recruiting function, and hiring managers who need a first round that holds up to being questioned.
- Browser voice
- Phone call
- Text chat
No facial analysis
Nothing about a face enters a score.
Every score is quoted
Each one cites the transcript line behind it.
No automated rejections
A named person clicks decline, with a reason.
Who it is for
Making an effective placement decision is hard
Most of what makes it hard is not the work — it is who interviewed well, who reminded you of someone, who you spoke to last. This takes the emotion out of the first round and puts the best fit for the role in front of you, with the evidence for it attached.
HR teams with no recruiting function
You own hiring on top of payroll, benefits and everything else. This runs the first round end to end and hands you a scorecard a hiring manager can read in five minutes.
Hiring managers who interview rarely
Interviewing well is a skill that decays between openings. The competencies, the follow-ups and the anchors are the same for every candidate, so the comparison holds up.
Anyone who has to defend a decision
Agencies, regulated employers, and teams hiring into New York or Illinois. Disclosure, consent, quoted evidence and an audit log for every access come as standard.
One role, everyone who applied
The whole pipeline as one list
Every candidate, what they scored, which channel they chose, and what still needs a person. Filter it, save the view, export it, or send two to eight of them to the comparison.
kestrel · candidates
How it works
Four steps, in this order
The numbering is the sequence a candidate actually moves through, from the application to the moment a person on your team makes the call.
Step 1
They apply on your careers page
Branded page or embeddable widget. The resume is parsed on upload; gaps, overlaps and inflated titles are flagged for the interviewer to probe, never used to reject.
kestrel · step 1
Step 2
They interview right away, their way
Voice in the browser, a phone call, WhatsApp, or text chat, in English. No account, no app, and no timer punishing someone who thinks before answering.
kestrel · step 2
Step 3
Every answer is scored against an anchor
25 to 40 minutes: resume walk-through with verification, STAR deep-dives with adaptive follow-ups, role scenarios, and consistency checks woven in. Each score cites the line it came from.
kestrel · step 3
Step 4
Your team decides
Stack-rank the shortlist, read the arithmetic behind any ranking, and advance or decline. Nobody is declined without a person clicking decline.
kestrel · step 4
Why it matters
Hiring is expensive, slow, and mostly unstructured
2×more predictive
Structured interviews predict job performance about twice as well as unstructured ones — validity .42 against .19.
Why every candidate gets the same competencies, the same rubric and the same anchors.
Source: Sackett et al. (2022) meta-analysis; Schmidt & Hunter (1998) found .51 against .38
60%never finish applying
Three in five workers have started a job application and abandoned it before submitting.
Why the interview starts immediately, in the browser, with no account and no app.
Source: iCIMS, 2025 State of Frontline Hiring Report
$4,129average cost per hire
The most-cited industry benchmark for what a single hire costs in direct recruiting spend alone.
Why the first round is worth doing properly rather than screening on a resume scan.
Source: SHRM Human Capital Benchmarking Report, fiscal year 2015 — dated, and almost certainly higher now
$1,500a day, per violation
New York City fines employers who use an automated hiring tool without an annual bias audit or candidate notice — and counts every single day the tool is used as a separate violation.
Why disclosure, consent and an audit log are built in rather than bolted on.
Source: NYC DCWP, Local Law 144 ($500 first violation, $500-$1,500 thereafter). Illinois AIVIA separately requires consent before AI video interviews.
What this will never do
The features we left out are the point.
Most AI interviewing tools score a candidate on how they look and sound. That is the part regulators are closing in on, and the part that is hardest to defend when someone asks why they were rejected.
No facial analysis
No video scoring, no micro-expressions, no gaze tracking. Nothing about a face enters a score.
No scoring on how you sound
An accent, a stammer or a flat affect cannot move a number. Where an answer is delivered unlike the same candidate's other answers, that prompts one more question and never a score.
No personality inference
No five-factor profile, no culture-fit index, no proxy for traits nobody agreed to be measured on.
Pre-send classifier · example
Blocked before it was asked
“Do you have kids at home who might make the weekend shifts difficult?”
Asked instead
“This role needs two weekends a month on site. Does that work for you?”
The candidate never hears the blocked draft, and the block is written to the audit log.
Calibration
Check the scoring against your own judgement
Score a handful of transcripts yourself and see where the model and your team disagree, competency by competency. A rubric nobody has checked is just a number with a label on it.
kestrel · calibration
Live demo
Two employers, five open roles
These careers pages are real and run the full interview. Apply with any name and interview by voice in your browser, by phone at +1 877 467 3541, or by chat.
Electronics manufacturing · Duluth, GA · English
ACI Atlanta careers
Engineering Support / Junior Manufacturing Engineer — CAD, DFM, yield investigations, and a reflow-failure scenario.
1 open role
Destination weddings · Cancun / Riviera Maya · Remote
WeddingPlan360 careers
Coordinator, content creator, product engineer and part-time sales — resort negotiation, content-ownership verification, shipping judgement and closing couples.
4 open roles
Compliance
Built to the strictest US standard, and for Mexico
- Disclosure and consent
- AI disclosure and consent before every interview; recording consent captured and logged. Written for Illinois AIVIA, NYC Local Law 144 and the Colorado AI Act posture.
- What is evaluated
- No facial analysis, no emotion analysis, no personality inference. Only words are scored. Delivery is compared with the same candidate's own baseline to decide where to ask for a specific, is disclosed to the candidate, and never reaches a score or a ranking.
- Data and retention
- Mexico LFPDPPP privacy notice, GDPR-ready deletion requests, 12-month default retention, per-tenant isolation at the query layer, and an audit log for every access.
- Integrations
- Greenhouse, Lever, Workable, BambooHR, Indeed and LinkedIn feeds, OCC Mundial, a REST API and Zapier-compatible webhooks.