Talk to us
Menu
ChaosHiring intelligenceClarity
ChaosOne record per candidateClarity
Bash · Hiring intelligence

Your ATS runs the pipeline.
Nothing runs the interview.

Bash does. One agent prepares it, sits in it, and verifies it.

Talk to usRead a real report
The problem

The interview loop, today.
And with Bash.

An agent for the interview process. Give it a role and a resume and it does the work a good recruiting team would do if it had unlimited time.

Pre-call
Researches the candidate and their employers. Writes the question plan. Talks to the candidate before the interview.
In the call
Sits beside the interviewer.
Post-call
Transcribes and analyses the conversation. Verifies every claimed skill against what was actually said.
Next round
Carries what it learned forward, so nothing is asked twice and nothing is dropped. Ranks the contenders.

A person makes the hire/no-hire call.

Today’s interview loop
RESUME ARRIVES“80% test coverage.” Nobody checks it.INTERVIEWForty minutes in, testing never comes up.DEBRIEF“Seemed strong.” Nobody can say on what.NEXT ROUNDAsks what round one already answered.
  1. Resume arrives“80% test coverage.” Nobody checks it.
  2. InterviewForty minutes in, testing never comes up.
  3. Debrief“Seemed strong.” Nobody can say on what.
  4. Next roundAsks what round one already answered.
  5. ↺ repeats
With Bash
next round · nothing asked twicePRE-CALL REPORTResearched. 83 questions. A plan.ROUND ZEROThe candidate’s words, first.INTERVIEW · CO-PILOTOne question at a time.YOU DECIDEPOST-CALL REPORTEvery claim checked.HANDOFFOpen items carry forward.COMPARESide by side, cited.
  1. Pre-call reportResearched. 83 questions. A plan.
  2. Round zeroThe candidate’s words, first.
  3. Interview · Co-pilotOne question at a time.
  4. You decide
  5. Post-call reportEvery claim checked.
  6. HandoffOpen items carry forward.↺ next round · nothing asked twice
  7. CompareSide by side, cited.
  8. ↺ repeats
Where it sits

Hiring intelligence sits between the resume arriving and the offer going out.

Job boards and the ATS bring resumes in. Offers follow a yes. The middle runs on calendar invites and memory.

Bash runs the middle.

BASHYOUR JOB BOARDS & ATSOFFER & ONBOARDINGSourceApplyScreendossier fromthe resumeRound zero8-min conversationInterviewCo-pilot +recordAssessproctored testHandoffcarries therecordCompareside byside, citedDecideyour call,evidence attachedOfferOnboard
  1. Your job boards & ATS
  2. Source
  3. Apply
  4. Bash
  5. Screendossier from the resume
  6. Round zero8-min conversation
  7. InterviewCo-pilot + record
  8. Assessproctored test
  9. Handoffcarries the record
  10. Compareside by side, cited
  11. Decideyour call, evidence attached
  12. Offer & onboarding
  13. Offer
  14. Onboard
THE RECORD, AS IT TRAVELS
A resume arrivesDossier: resume + employer researchThe candidate’s own wordsTranscript + what was coveredTest score · integrity kept separateStill unknown → carries to next roundSide by side, every mark citedYour call, with the record attachedOffer
How it works

One agent. Every stage of the interview.

Six things it does on its own, for any role.

01 · Pre-call report

It builds the dossier.

Every employer researched, skills mapped to the role, and the plan: what to probe, what to ask, how long.

02 · Round zero

It gives the candidate a round zero.

Eight minutes, chat or voice, about work they actually did. Not scored.

03 · Co-pilot

It sits beside the interviewer.

One question at a time, what to listen for, a nudge only when it matters.

04 · Post-call report

It writes the report.

A minute after the call: a match tier, a confidence score, every claim checked against what was said.

05 · Handoff

It hands the next round what it learned.

Verified, mixed, still open. Anything never asked carries to the next round.

06 · Compare

It ranks the contenders.

Side by side, every mark cited to its report. You make the call.

Every line traces to something the candidate wrote, said or did.
Pre-call

Pre-call report

Built the moment a resume lands on a role. The interview, planned before anyone opens the resume.

  • Career timeline. A research brief on every past employer.
  • Skills against the role. A match percentage per skill, and the gaps.
  • Verification matrix. Resume claim, public record, verdict. Per skill.
  • Around 80 questions. Each with an ideal answer, a bad one and a follow-up.
  • A timed plan. Must cover, should cover, minutes per topic.
  • Shared in one click. Expiring link or PDF.
Resume · PDFCandidate ABackend Developer
Pre-call report · building
Career timeline rebuilt
3 employers researched
Skills mapped to the role
83 questions written
Questions ready83ideal, bad, follow-up
Critical gaps5with why each matters
Plan105minmust cover 95
Verification matrix · one of 24 rows
PostgreSQLResume: stored procedures, temporary tablesGitHub: no portfolio evidenceExplore
Real pre-call report, anonymised. Counts and the verification row are verbatim.
Round zero · as the candidate sees it
Candidate A
Backend Developer · interview scheduled Thursday 10:00
FORMAT
Chat · 8 min
YOUR LOGOfrom: your hiring team · to: the candidate
Before Thursday: eight minutes on your work

Your interview for Backend Developer is booked for Thursday at 10:00. Before then, tell us about the work you are proudest of, in your own words. Eight minutes by chat, or twelve by voice, whenever suits you.

Optional. Declining changes nothing about your interview.

Start the conversation
Before we start
I understand this conversation helps prepare my interview and is shared with the hiring team.
I agree to my answers being stored for this hiring process.
Optional: record my voice answers.
Optional: keep my answers beyond this process.
Continue
ROUND ZERO · BACKEND DEVELOPERCHAT · RESUMES IF YOU CLOSE THE TAB
0:42
CONVERSATION COMPLETE · 7 MIN 40 SNO SCORE · NO VERDICT
Pre-call report · Candidate A
Initial→Enriched

Rebuilt with the candidate's own words, before anyone interviews them. The interviewer opens on what they said, not only on what the resume claimed.

Questions, ideal answers and the focus plan are regenerated. Nothing is scored.

The sequence as the product runs it. Email and question wording are representative, not verbatim; the candidate's answers are not shown.
Before the call

Round zero

An optional conversation between Bash and the candidate, before anyone interviews them.

  • Sent on its own. A private link when the interview is scheduled, under your logo.
  • Their questions. Written from their resume and the role. Chat or voice. Resumes if they close the tab.
  • Guard rails. Two consent ticks. No salary or process questions. Protected topics screened out.
  • No score, no verdict. Their answers go into the pre-call report.
Candidate A · Technical roundLive 14:20 · Q3 of 5
Transcript · candidate speaking
Probe furtherThin on multi-region. Ask for a concrete trade-off.→ e.g. fail-open vs fail-closed under partition
CoveredIncident response. Move on.→ Named the escalation path and the fix
Topic coverage
System design
40%
Incident response
done
Automated testing
next
Nudges and topics as the product gives them. The candidate’s words are not shown.
During the call

Co-pilot

A second screen for the interviewer, live five minutes before the call. Nothing to install. The candidate never sees it.

  • One question at a time. With what good, weak and scripted answers sound like.
  • Coverage as you go. Each topic pending, done or skipped.
  • Six nudges, only when they apply. Probe further. Covered. Time. Off topic. Culture. AI pattern.
  • The interviewer stays in charge. Skip, next, more questions. Phone companion by QR.
  • The call is captured. A notetaker records and transcribes.
Post-call

Post-call report

Ready under a minute after the transcript. A recommendation, a confidence score, and the evidence for both.

  • Verification. Every claimed skill confirmed, weak or not discussed, with the line that decided it.
  • Moments. The ten exchanges that mattered, scored, with the clip.
  • Timeline. The recording, by topic. Click to jump.
  • Summary. Strengths, concerns, talk ratio, and what was expected but never discussed.
  • Behavioural signals. Fourteen, each with the quote behind it. Not a personality test.
  • Transcript. Every answer rated. Interviewer feedback per round. Share or PDF.
Post-call report · Candidate A · Round 2claimed vs shown
Claimed · resume + JD4/10Node.js
Shown · interview8/10Under-soldDemonstrated excellent understanding of the event loop and task priority. 10/10
Claimed · resume + JD3/10MongoDB
Shown · interview8/10Under-soldClaimed at 3 on the resume. Scored 8 in the interview.
Claimed · resume + JD—/10Automated testing
Shown · interview—/10Not discussedExpected by the job description. Never asked. Carries to next round.
Recommendation · resolves as the skills are checked
Partial Match
65%confidence
6/24skills verified
Scores, tiers and the quoted moment are verbatim from a real report, anonymised.
Round 1 · Technical · Tuesday
MongoDBVerified
PostgreSQLMixed signals
Express.jsStill open
PostmanStill open
Round 2 · Thursday · opens on this
carries from round 1
carries from round 1
Established on TuesdayMongoDB verified · PostgreSQL mixed
Express.js→ carries to next round
Postman→ carries to next round
Real record, anonymised. The next interviewer opens on the right-hand card.
Between rounds

Handoff

The candidate record, updated after every round. The next interviewer opens on this, not the resume.

  • What changed. Each skill verified, mixed or still open, by round.
  • Evidence so far. Score before, score after, status, round.
  • Still unknown. Anything never asked, marked carries to next round.
  • One place. The whole journey and the team’s notes.
Deciding

Compare

Two or three finalists on one page, ranked, every claim linked to its report.

  • Ranked. By match tier, then assessment score. Not-yet-interviewed kept apart.
  • Per criterion. Stronger, comparable or no evidence, cited to the round.
  • What separates them. Worth a look. What we don’t know. A next step for each.
  • On record. Stored scores and dates, not AI-authored. PDF.
From the job descriptionDatabase scalability and optimisation
StrongerPostgreSQL and MongoDB tuning, indexing strategies.Post-call report · Round 1
ComparableExecution plans and data-modelling trade-offs.Post-call report · Round 1
From the job descriptionNode.js and Express.js proficiency
ComparableRESTful APIs with Express in production.Post-call report · Round 1
StrongerEvent loop, non-blocking I/O, worker threads.Post-call report · Round 1
From the job descriptionAutomated testing and quality assurance
No evidenceNot weaker. Unknown; the round did not assess it.Post-call report · Round 1
No evidenceNot weaker. Unknown; the round did not cover it.Post-call report · Round 1
Rank 1Candidate ALikely Match · 75% confidence
Rank 2Candidate BLikely Match · 73% confidence
What we don’t knowBoth: automated testing. Next step for both: a practical exercise on test implementation.
Real comparison, anonymised. Marks and reasons are verbatim.
Coding · Median Window Stream · 30 min
Hidden tests0/4
Two grading passesagreed
Integrity · kept separate5.0Pastes 0 · Copies 0 · Tab switches 1Not penalised automatically. Recorded for your review.
Integrity timeline · 73 mintab switch ×130-min limitsubmitted · 73 min
Real assessment report, anonymised. Score and integrity are separate records.
Any stage

Assessments

Coding, MCQ, case study or project. Generated from the job description, or imported from your own test.

  • Built in minutes. From topics, the JD, or a PDF, DOCX or text.
  • Tests that run. Visible as they go. Hidden on submit.
  • AI, on your terms. Per question, an assistant for the candidate, within a token budget you set.
  • Two graders agree. Correctness 80, code quality 20. Two passes before a score.
  • Integrity, kept separate. Pastes, tab switches, timing: shown, not penalised. The candidate is told.
  • Work style, ungraded. Optional. Bands, not scores.
Three doors

If you interview.

A plan you didn’t write. A screen that says when to move on. No notes. Your job becomes the conversation.

If you recruit.

A dossier the moment a resume lands. A round zero you didn’t schedule. Reports the panel reads before they ask you. The chasing goes away.

If you own the hire.

The same evidence for every candidate, whoever interviewed them. A comparison that cites its sources. A record you can defend six months later.

You may already have part of this
TodayA notetaker

Records the call and summarises it.

With BashThe same recording, checked

Bash records too, and knows what the resume claimed and what the job needs. The report checks one against the other.

TodayAn assessment platform

Tests in a sandbox.

With BashThe test, next to the interview

Bash runs assessments too, and reads the result beside what was said in the room.

TodayATS scorecards

The same form for everyone, filled in from memory, late.

With BashThe scorecard, from the transcript

Filled in from what was said, a minute after the call, with the line behind each mark.

Proof

Inside a 90-person AI company

26people interviewing across 14 roles
142interviews recorded · 110 hours · 56 min average
114Pre-call reports · 117 Post-call reports
69/133scored reports came back Not a Match

Interviewing runs on gut feel almost everywhere, and we were no exception. Bash flagged gaps in people we’d already screened in and pointed at the line behind each one. Our debriefs now run on evidence.

— CEO, Wobot AI
That last number matters. All 69 had passed screening. The report said no anyway, and showed where. A tool built to flatter does not say no half the time.
Before you book
Does it decide who to hire?No. It gives a match tier, a confidence score and the evidence behind them. It never rejects anyone. A person decides.
Does it screen applicants?No. Bash works on candidates you have already chosen to interview.
Does the candidate know?They see a notetaker join, as with any recording bot. Round zero is consented to and can be declined. Co-pilot is invisible to them. No camera, no eye tracking. In assessments, pastes and tab switches are recorded, shown to you, and disclosed to the candidate.
Is the behavioural section a personality test?No. Behavioural signals are patterns from one conversation, each with its quote. The work-style questionnaire is optional, ungraded, and cannot filter anyone.
How fast is the report?Under a minute after the transcript.
What languages?English today.
We already use an assessment platform or a notetaker.Keep it. Nothing has to change on day one.
Where does the data live? Is it secure?SOC 1 and 2, ISO 27001 and GDPR readiness are in progress with an external audit firm. None is complete yet, and we won’t claim one until it is. Reports are visible only to the roles you assign. Ask for the security overview.
How much does it cost?Talk to us.
Works with the tools you already useGreenhouseWorkdayZoho RecruitWorkableRecruiteeGoogle CalendarOutlook CalendarZoomGoogle MeetMicrosoft Teams

Hire the right people, with confidence.

Bring a resume, a job description and a recording of one of your interviews. We’ll show you what the room missed.

Talk to us