Your AI hiring team, with the evidence attached.
Pre-launch · early access opening to a small number of teams
YoHe reads every applicant's documents against the job you defined and returns named strengths, named gaps, and the quote behind each one. You still make the call — with the reasoning in front of you.
Workspace
PipelineJobsInboxSearchSenior Backend Engineer · 84 applicants
Where to start reading.
Strong evidence
18
21% of pool
Possible match
23
27% of pool
Low signal
43
51% of pool
Synthetic data. No real candidate appears anywhere on this site.
Four hundred applications, one afternoon, and no way to defend a single decision.
Keyword filters throw away good people for the wrong vocabulary. Reading everything by hand does not scale past a few dozen. Either way, when someone asks why a candidate was passed over, the honest answer is usually a shrug.
From a structured job to a defensible shortlist.
Intake, assessment, triage, and interview prep in one place — each step showing its work.
// 001
Assessment with the evidence attached
Each requirement comes back met or not met, with the sentence from the candidate's own documents that supports it.
Strong evidence against 9 of 12 requirements.
- Backend engineering6 years
“Led backend architecture for a multi-tenant billing platform serving 40+ enterprise accounts.”
resume.pdf · p.1
- TypeScript and Node.js in production5 years
“Rebuilt the ingestion service in TypeScript; cut p99 latency from 2.4s to 310ms.”
resume.pdf · p.2
- Kubernetesnot evidenced
Your resume mentions container orchestration. Which parts did you own, and what broke?
// 002
Jobs defined as requirements, not prose
Turn a job description into weighted, checkable requirements. The model scores against those, not against a vibe.
- MustBackend services in production6 yrs
- MustTypeScript and Node.js5 yrs
- MustPostgreSQL at scale4 yrs
- NiceKubernetes—
- NiceLed a team or workstream2 yrs
// 003
Triage that sorts, never rejects
Applicants are banded by evidence strength so you know where to start reading. Nothing is closed out without a person.
Pool · 84 applicants
No candidate closed out
- Strong evidence18
- Possible match23
- Low signal43
// 004
Interview questions from the actual gaps
Where evidence is missing, YoHe drafts the question that would settle it, so the screen is spent on what you do not know yet.
Unevidenced requirement
Kubernetes
Your resume mentions container orchestration. Which parts did you own, and what broke?
A score you cannot open is a guess with a number on it.
YoHe is built so that every claim about a candidate can be traced back to the words they wrote.
- Quote, document, page// 001
- Findings cite the exact passage. If the passage is not there, the requirement is marked unevidenced rather than estimated.
- Gaps are stated, not hidden// 002
- What the documents do not show is listed as plainly as what they do, so a thin application is visible instead of averaged away.
- Disagreeing is a first-class action// 003
- Override any finding. Your correction is kept next to the original and travels with the candidate record.
- Readable months later// 004
- The record holds the job version, the requirements in force, the model output, and every human edit, in order.
Strong evidence against 9 of 12 requirements.
- Backend engineering6 years
“Led backend architecture for a multi-tenant billing platform serving 40+ enterprise accounts.”
resume.pdf · p.1
- TypeScript and Node.js in production5 years
“Rebuilt the ingestion service in TypeScript; cut p99 latency from 2.4s to 310ms.”
resume.pdf · p.2
- Kubernetesnot evidenced
Your resume mentions container orchestration. Which parts did you own, and what broke?
Synthetic example · not a real candidate
Four steps, each one auditable.
The loop is short on purpose. Every step writes to a record you can read back months later.
// 01
Define the job as requirements
Paste the description. YoHe proposes weighted requirements; you edit, add, and remove until it matches the role you are actually hiring for.
// 02
Collect applications in one inbox
Resumes, cover letters, portfolios, and forms land together. Documents are parsed once and kept intact alongside their extraction.
// 03
Read the evidence, not a score
Each candidate returns requirement-by-requirement findings with the supporting quote and its source document and page.
// 04
Decide, and leave a record
Advance, hold, or decline with a reason. The decision, who made it, and what was on screen at the time are written to the audit log.
Four things YoHe will not do.
These are product constraints, not settings. They are not configurable away.
// 001
No auto-reject
Every rejection is made and sent by a person.
// 002
No inference
Age, ethnicity, gender, religion, disability, and nationality are excluded and never inferred.
// 003
No face or voice scoring
No facial analysis, emotion recognition, or personality prediction. Ever.
// 004
No silent AI
Anything the model wrote is labelled as such and stays editable.
Designed for the part where someone asks you to justify it.
Hiring decisions get challenged — by candidates, by regulators, by your own team. The system should make that easy to answer.
Human in the loop, structurally
There is no code path that rejects, ranks-out, or hires a candidate without a person acting. Automation stops at the recommendation.
Protected characteristics excluded
Names and photographs can be hidden during assessment. Protected characteristics are never used as matching signals and the system does not attempt to infer them.
Criteria you control
Requirements and weights are yours. You can see exactly which requirement drove which finding, and change it.
An audit log you can export
Every assessment, edit, override, and decision is timestamped and attributed, and leaves the system in a format you can hand to counsel.
- AIAssessed against requirements v312 findings
- YouOverrode finding: KubernetesMarked evidenced
- YouAdvanced to first interviewReason recorded
- AIDrafted interview questions4 questions · edited
Synthetic example · every row timestamped and attributed
Ask the pipeline a question in your own words.
Instead of building a filter, describe what you are looking for. Every answer comes back with the candidates and the evidence behind the match.
Matches with evidence
- Candidate 1187Senior Backend Engineer · 4 documents
“Owned the Stripe migration from contract to launch.”
- Candidate 0942Senior Backend Engineer · 3 documents
“Built payouts and reconciliation for a marketplace.”
- Candidate 1044Platform Engineer · 2 documents
“Integrated a card processor; no end-to-end ownership stated.”
- Who has shipped a payments integration end to end?
- Which shortlisted candidates have no evidence of team leadership?
- Show me people strong on the top three requirements but light on the fourth.
- What should I ask candidate 1187 about their gap in distributed systems?
The things people ask first.
- Does YoHe reject candidates automatically?
- No. YoHe assesses and bands applicants so you know where to start reading. Advancing, holding, and declining are all actions a person takes, and each one is recorded against the person who took it.
- Can it tell me who the best candidate is?
- It can tell you which requirements each candidate has evidence for, and which they do not. It does not rank people into a single best-to-worst order, because the trade-offs between requirements belong to your team, not to a model.
- What about bias?
- No tool can make hiring unbiased, and we will not claim ours does. What YoHe does is narrower and checkable: it excludes protected characteristics from matching, never infers them, lets you hide names and photographs during assessment, and shows the evidence behind every finding so a biased conclusion is visible rather than buried in a score.
- Where does candidate data go?
- Candidate documents are processed to produce assessments for your team and are not used to train models. Data handling is described in the privacy policy, and the details will be documented in full before the product is generally available.
- Which AI models does it use?
- YoHe is model-agnostic by design and runs on frontier models from more than one provider. The provider and model version used for an assessment are recorded with the assessment.
- When can I use it?
- YoHe is in development. Early access is opening to a small number of in-house hiring teams. Requesting access puts you in the queue for that group.
Get started in a minute
Hiring, on the record
Tell us what you are hiring for. We will get in touch as early-access places open up.
> Early access
Try YoHe first.
Tell us about your hiring and we will get in touch as places open up. No commitment, and no sales sequence.