One score per candidate. Every input behind it, on the record.
You pick the skills for the job and give each one a weightage. You set up the job description. Answers are scored against both, and follow-up questions adapt to what the candidate says. The score comes back out of 100, built from the transcript, the code that ran, the test cases that passed and the resume text. This page covers all of it, plus what we do not score and where a person takes over.
30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days.
- One score, 0 to 100
- Your skills, your weightage, your JD
- Evidence attached to every score
- A person makes the decision
A readiness score measures a candidate against the job, not against other people.
Most arguments about AI scoring start because nobody says what the number is. Here it is. Four things define a readiness score, and you can see all four in the product.
It is out of 100.
One integer, shown as a ring next to the role and the round. No second hidden number, no letter grade, no star rating.
It scores against your skills and your JD.
The skills you picked, the weightage you gave each one, and the job description you set up. A candidate's score does not move because someone else in the pipeline did better. It is not a percentile.
It is comparable across rounds.
Resume screening, AI interview, human-led interview, prompt assessment and coding test all score against the same skills for that job. So a resume score and an interview score sit on the same axis.
It rolls up without changing meaning.
The same number adds up. Per candidate for an employer, per student for practice, and to a batch readiness score for a campus.
The bands the product reports in.
Readiness distributionWhen a report shows a group rather than one candidate, scores are grouped into four bands. A placement cell sees the same four bands on a readiness distribution.
Top band in readiness reporting
The largest band in most cohorts
Coaching band
Furthest from the bar
One line sits tighter than the bands. A skill bar on a reportcard renders green at 70 and above, amber below it. That is a reading aid on one skill, not a pass mark on the candidate.
Three pillars, and how they add up.
Every round scores technical accuracy, communication and problem solving. What changes by round type is the evidence each one is read from.
Technical accuracy
Did the work run, and was it good work.
On a coding test this is decided by execution. The submission runs in a sandbox against hidden test cases, and the panel reports X of Y passed, the runtime of each case in milliseconds, and a complexity verdict. On an interview round it is read from the transcript, against the technical skills on the job.
Communication
Could they explain the thing they built.
Scored from the speaker-separated transcript. Is the answer structured, is the claim supported, does a complex idea survive being explained simply. The input is text, so this is not a read on accent or delivery.
Problem solving
The route, not just the destination.
This reads the sequence: how edge cases were handled, how a failing case was debugged, how the approach changed after a follow-up. A candidate who fixes a wrong assumption tells you more than one who pastes a memorised solution. On a prompt assessment it is iteration strategy, one of the 8 dimensions.
How the weights are set. You set them.
- 01You pick the skills for the job and give each one a weightage. Skills can be suggested for you, split into must-have and good-to-have, and you review them before they save.
- 02You set up the job description. Paste your own, upload a file, or have one drafted and accept it.
- 03Answers are scored against those skills and that JD. Follow-up questions adapt to what the candidate says.
- 04The score comes back mapped to your skills, one number each, plus one overall out of 100.
There is no global weight we publish for you, because a backend role and a customer-facing role do not weigh the same things. What is fixed is that the skills and the weightage are set once, before candidates arrive, and applied to everyone in that round. Each skill gets its own number on the reportcard, next to a verdict line in plain words.
Set skills and weightageFour evidence sources. Nothing scored that is not one of them.
Six months later, someone asks why a candidate was moved on. Every number on a reportcard resolves to one of four things, and all four sit on the report rather than in a separate system.
The transcript
Speaker separated and searchable, with All, Candidate and Interviewer filters, sitting next to the recording. Every communication and technical score in a spoken round traces back to a line in here.
The code
What the candidate wrote in the editor, captured as it happened, with the language they picked and any custom stdin they ran. The approach itself, not a description of it.
The test cases
Real execution in a sandbox against hidden test cases. Reported as X of Y passed, the runtime of each case in milliseconds, and a complexity verdict. It either ran or it did not.
The resume
For a resume screening round: skills, experience and education read against the role, no interview. The resume stays on the report as its own tab.
Reports are never metered, so you do not ration the evidence. Each one is share-ready and downloads as a PDF, which matters when the person who needs the reasoning is a hiring manager, a client or a regulator without a login. How coding tests are executed and scored
This reportcard is the argument.
Two screens. The first is the round while the candidate is still in it. The second is what a reviewer opens afterwards, where the skill breakdown, the recording, the transcript and the integrity status sit on the same page as the score.
The round, while it is happening.
A sandboxed problem with a difficulty tag and topic tags, a language picker, a line-numbered editor and optional custom stdin. The results rail reports 5 of 5 passed with the runtime of every case, then a complexity verdict of O(n·k), marked optimal. The header carries the timer and a Proctored badge counting tab switches.
The score, with its working shown.
The same candidate as a reportcard: 84 overall in a ring, a Strong fit for role chip, the round number and name, and a verdict line. Under it, the skill breakdown labelled Scored from transcript and code: Flask and REST APIs 92, System design 81, Code quality 78, Communication 64. The rail holds the recording, the transcript and the malpractice status.
Product screens shown with sample data. Candidate names and companies inside Expert Hire mockups are invented.
One candidate, from first line of code to a score of 84.
The same candidate as the screens above, walked end to end. The names and numbers are sample data from our mockups. The mechanics are real.
The job and the round are set up first.
A Python Software engineer role working in Flask. The recruiter picks the skills, weights Flask and REST APIs highest, and sets up the JD. Then Round 2, Tech Screening: a coding test on Coin Change, tagged Hard, with proctoring on.
The candidate works, and the work is captured.
The candidate picks Python, writes a bottom-up solution, runs it against custom input, then submits. It executes against five hidden cases: two samples, an impossible amount, a large amount and a single coin. The Proctored badge counts three tab switches.
Execution decides the technical score.
All five cases pass, runtimes 8ms to 31ms, approach graded O(n·k) and marked optimal. That is what the sandbox returned, not a model's opinion. Code quality is read from the submission itself and lands lower, at 78.
The transcript decides the spoken scores.
The recording runs 16:41 and produces a 96 line transcript. System design scores 81. Communication scores 64, the weakest number on the card, because the explanation lagged behind the execution. Both point at lines a reviewer can go and read.
The weighted skills produce one number.
The skill scores roll up to 84 out of 100, a Strong fit for role chip, and a verdict line: strong Flask expertise and acceptable must-have skill scores. The three tab switches are reported as malpractice status, not folded into the score.
A person decides what happens next.
The reportcard lands in a reviewer's queue. They read the 64 on communication, listen to two minutes of the recording, and shortlist the candidate anyway because the role is backend. That decision is an action in the workspace, so the record shows both.
The same round as a ledger.
Round 2 · Tech ScreeningLeft to right: the skill, the number, and where the number came from. Nothing in the middle column exists without something in the right column.
| Skill | Score | Traces back to |
|---|---|---|
| Flask and REST APIs | 92 | Sandbox execution and transcript |
| System design | 81 | Transcript, follow-up questions |
| Code quality | 78 | The submitted code itself |
| Communication | 64 | 96 line speaker-separated transcript |
| Overall readiness | 84 | The weighted skills and the JD, applied to every candidate in this round |
Communication at 64 renders amber because it sits under 70. It does not fail the candidate. It points the reviewer at the part of the recording worth listening to.
Scored on what they said, wrote and built. Not on a face scan.
Regulated buyers ask this first. Most vendors answer it last. Here are the exclusions, stated flatly.
Accent, and how someone sounds
Spoken rounds are scored from the transcript, which is text. The question is whether the answer was structured and supported, not how the delivery sounded.
Facial expression and appearance
No face scan feeds a score. No attractiveness signal, no expression reading. The skill breakdown is labelled Scored from transcript and code, and that label is literal.
Personality inferred from video
We do not infer traits from a recording and score the inference. Psychometric Test, our psychometric profile, is a separate product a job seeker chooses to take. It does not feed an employer's reportcard.
Name, gender and age as inputs
Identifying details on a resume can be masked, both in the Chrome extension before a resume is shared and server-side before a report leaves the pipeline. With masking on, those fields never reach the scoring.
What is scored, in full.
- What the candidate said, from the transcript
- What the candidate wrote, from the editor
- What actually ran, from the test cases
- Skills, experience and education against the role
That is the whole list. If an input is not on it, it is not producing the readiness score.
Two honest carve-outs about video.
First, video is analysed after an interview, for integrity only. Live signals and post-interview analysis look for known AI assistant overlays, signs of impersonation and tab switching. They are reported on the card as a malpractice status and go to a human reviewer. They are not a scoring dimension, and a flag is never an automatic verdict.
Second, job-seeker practice mock interviews do give the candidate delivery coaching back, which can name something like posture. That is practice feedback a student asked for about their own attempt. An employer never sees it.
The score recommends. A person decides, and the record shows both.
We do not auto-reject anyone on our own judgment. That sentence is worth little on its own, so here are the four points where a person sits in the flow, and what the trail shows when they disagree with the score.
Before the decision
Completed rounds land in an Awaiting your review queue on the recruiter dashboard, next to interviews this week and average time to shortlist. A score arriving is a prompt to look.
During the interview
A recruiter can silent-listen to a live AI interview without interrupting it, then take over in real time. The human is not stuck reviewing after the fact.
At the decision
Shortlisting, rejecting and moving a candidate are recruiter actions in the workspace. The score is a recommendation with evidence attached.
After the decision
Audit logs cover every action, and the reportcard keeps the score, the breakdown, the recording and the transcript. A review months later sees both.
What an override actually is here.
There is no override screen, because the score never had the authority to be overridden. A reviewer who disagrees with an 84 shortlists or rejects that candidate the same way as any other. That is the decision of record.
Audit logs cover every action, so the disagreement is part of the trail. The score stays on the reportcard, unchanged and still traceable, next to the human decision that went the other way.
Integrity signals feed review, not rejection.
We detect known AI assistant overlays, signs of impersonation and tab switching during live rounds, plus video analysis afterwards. Each signal goes to a human reviewer with the evidence. A flag is never an automatic verdict, and a clean round is reported too: the card reads None, no tab switches flagged.
Rounds do have a promotion threshold, and auto-promote and auto-reject toggles. Those are settings you choose and own, they are off unless you turn them on, and they run on the score. A malpractice flag never rejects anyone.
How malpractice detection worksFive round types. Different evidence, the same skills.
A pipeline is built from rounds, and every stage is one of five types. They do not measure the same thing the same way. What they share is the skills and the JD they score against, which is what makes a resume score and a coding score comparable for one candidate.
Resume Screening
Scored on: Skills, experience and education against the role
The only mode with no interview. The AI reads the resume against the job and returns a score. Runs in bulk, sorts, and supports Shortlist top and CSV export.
AI Interview
Scored on: Communication, confidence and technical depth
Ethan, our AI interviewer, runs a voice round 24/7 and in the candidate's language. Follow-up questions adapt to the answers, so the transcript is what keeps the round auditable, not the question list.
Human-Led Interview
Scored on: The same skills and JD as an AI round
A live recruiter-run meeting, hosted inside Expert Hire or on Zoom, Meet or Teams with a notetaker in the room. Recorded, transcribed and scored on the same reportcard as an AI round.
AI Prompt Assessment
Scored on: 8 dimensions including prompt quality and iteration strategy
Candidates write prompts for real tasks. Sorted by category, technical or non-technical, and level, with a set number of attempts per question, so iteration is part of what is measured.
Coding Test
Scored on: Correctness, efficiency and code quality
The least subjective mode. Code runs in a sandbox against hidden cases, reporting pass or fail per case with runtime, plus a complexity verdict. Execution decides correctness, not a model reading the code.
Inside an AI Interview, pick one of eight types.
Time limit per typeThe five above are the modes a round can run in. When the mode is AI Interview, you also pick the type. Coding and Task give the candidate a code editor. System Design gives a whiteboard. The rest are question and answer.
Technical coding assessment with programming questions and code editor
General Q&A interview covering broad topics and overall fit
Assessment of communication skills, clarity, and articulation
Logical reasoning, analytical thinking, and aptitude assessment
Architecture and system design discussion with whiteboard support
Interview for Faculty Recruitment at Universities
Human resources interview focusing on behavioral questions and cultural fit
Short focused coding task, single challenge plus 2-3 follow-ups, 10 minutes
What holds across all five.
AI and human rounds mix in the same pipeline. Rounds are numbered on the job, so a reportcard says which stage it came from. Proctoring is a modifier, not a round type: it adds an integrity status at no extra credits, and it does not change how the score is worked out.
Reports cost nothing. You never pay per unit for the evidence behind a decision.
What we do about bias, and what we have not done yet.
Most vendors answer this with four checkmarks and no evidence. The useful version has two columns. The second one tells you whether to believe the first.
What is in place today
The same skills and weightage for everyone in a round
Both are set before candidates arrive and applied identically. No per-candidate weighting, no retuning mid-round.
Identifying details can be masked
The Chrome extension masks PII on a resume before it is shared, and reports get server-side resume masking before they leave the pipeline. Masking counts as a resume scan, 1 credit.
Structured formats over free-form judgment
Every candidate in a round gets the same round type, the same question set and the same evidence requirement. That removes the drift that makes unstructured interviews hard to defend.
Reviewable records and exports
Scores tie back to transcript and code. Audit logs cover every action. Hiring data exports in one click for a regulatory review, an internal audit or legal discovery.
What we are not claiming
We have not published an independent bias audit of our models.
We will not imply we have. NYC Local Law 144 requires an independent third party to run that audit, and requires the employer using the tool to commission it. Doing it ourselves would not satisfy the law.
We do not sell a bias-audit product.
We reduce bias inputs at the source and hand you the records your own audit needs. If you want an auditor, we can introduce you to independent audit partners.
SOC 2 and ISO are in progress, not done.
Live today: encryption in transit and at rest, role-based access control, audit logs, single-tenant separation, and GDPR readiness with DPAs and right to erasure. The certifications are in progress. SSO and SAML are on the roadmap.
Take this page to your own jurisdiction.
What you have to show differs by where you hire, so we publish a guide per jurisdiction. The security measures behind all of it are documented in full.
Questions about the score, answered.
What buyers, auditors and candidates ask before they trust a number.
Still have a question? Book a demo and we will open a real reportcard with you.
Run a round and read the reportcard yourself.
Pick the skills, set the weightage, set up the JD, and run one round. Then open the report and go looking for the evidence behind the number. The trial has enough credits for that.
30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days. Reports are never metered.