HumsanaFor founders and HR teams
Hiring that finds
overlooked candidates
More than three hundred people can apply to one role, and nobody reads them all. Humsana reads every one against a rubric you approve, quotes the evidence, and records every decision.
Start with one roleSee how it reads
Nothing is read until you approve the rubric.
- ARAsha R.4 of 5 criteria carry evidence
- DKDaniel K.Follow-up needed
- MSMeera S.5 of 5 criteria carry evidence
- TOTomas O.1 criterion contradicted
- LWLena W.Follow-up needed
Asha R.
The problem
The pool got bigger. The reading got thinner.
At a company of under 250 people, one role now draws 312 applications and 11.5% of them are qualified. Reading the pile properly is not a job anyone has time for, so it gets filtered by title and keyword, or handed to a tool that answers with a summary nobody can check.
Applications for one role
312Qualified for the role
36Two hundred and seventy-six people are filtered on a word. Some of them could do the work and did not describe themselves in your language, and the filter cannot tell the difference.
Rules, or a confident guess
Keyword filters work like an ATS. A model instead writes a fluent paragraph nobody can trace back to the CV. Either way the reasons leave no trace.
Written with a model
29.3% of job seekers used AI to write or tailor a CV in the past year, up from 17.3%. Around 1% of real CVs now carry hidden text aimed at the screening model.
Duties that arrived with the tools
Automated hiring now carries duties on records, transparency and human oversight. A trail nobody can verify does not answer them, and this page does not claim otherwise.
Figures: Employ 2026 · iHire 2025 · USENIX 2026
How it works
Two ways in. One approved rubric.
Whether the applications are already in a folder or the role is still being written, the work runs in the same order and ends in the same place.
You already have applications
- 1Point Humsana at the job description you posted.
- 2Answer a short set of questions: the work to be done, the abilities required, how the person must operate, what would help the company grow.
- 3Humsana reads the pool. Names and identifying lines come off first, and text addressed to the reviewer is recorded as an attempt, never obeyed.
- 4Each criterion comes back evidenced, not evidenced or contradicted, with the passage behind it.
You are writing the role
- 1Answer the same questions about the role before it is posted.
- 2Humsana drafts the rubric from your answers, in your words rather than a template.
- 3Approve it, then take applications by form, sheet, CSV or mailbox. The rubric is fixed for that role.
- 4Read the evidence per criterion and decide. A gap becomes a question; only a contradicted criterion can end an application.
The record
Every decision comes with a receipt.
Open any candidate and see what was proposed, the evidence it rested on, the rule that checked it, and what happened. Pick a case.
Transparent. Every criterion, every quote and every rule that fired is on the page, in the order it happened.
Authentic. The record is signed. Change one character and the check fails, without an account and without asking us.
Bias at the floor. Names come off before the read, no applicant is sorted against another, and a missing answer becomes a question.
The authority engine came first. It was built to check what an AI agent is about to do against the authority it was given, and to allow, hold or refuse it before the action happens. Hiring is where that check earns its place.
Separation of duties
The AI reads. Fixed rules decide.
An agent should not police itself. Humsana separates the two jobs, so the part that understands messy CVs is never the part that approves an action.
What we will not do
Names come off first
Contact details and identifying lines are removed before anything reads an application.
No scores, no ranking
Each person is read against your rubric, never sorted against each other.
A person decides
A shortlist is a recommendation. The hiring decision is always yours.
Signed and checkable
Anyone can verify a record without an account.
How it is built
Four parts, one system.
Each one does a different job, and none of them decides who gets hired.
-
Company Hiring Brain
How work gets done here, what good looks like, and what someone must already know. Written once, approved once, and read by every role after that.
-
Evidence engine
Every claim carries the passage it came from, what the passage establishes, and what it does not. Missing evidence is a question, never a mark against anyone.
-
Open questions
Where a CV does not settle something that matters, the reading says so and names the question that would settle it. Humsana does not contact candidates; a person decides what to ask and sends it.
-
Authorization and verification
Every action is checked against what you granted before it happens, and the check is recorded either way. A signature proves the record was not altered; it does not prove the decision was right.
Questions
What people ask before signing up.
Does this replace our ATS?
No. It sits beside it. Applications arrive from a form, a sheet, a CSV export or a forwarded mailbox, and the shortlist goes back to wherever your team already works.
Can it reject someone for a missing answer?
No. An application can only end on a criterion the evidence contradicts. A gap becomes a follow-up question to the candidate, not a rejection.
What happens when a CV contains instructions aimed at the reviewer?
The text is read as data. The attempt is recorded with the words it matched, the rubric cannot change, and an advance is refused and held until a person has read the text and said so on the record.
Does it rank candidates against each other?
No. There is no score and no league table. Each person is read against the same rubric, and the record shows what each one carried.
Who sees the applications?
Your team, and a model that receives redacted text rather than the file. Names and contact details do not leave the system. The provider and its terms are named in the privacy notice.
Is the AI making the hiring decision?
No. It reads and proposes. A fixed set of rules checks the proposal against your approved rubric and can allow, hold or refuse it. A person makes the decision, and the record names them.
Run one role.
See what it reads.
Fifteen minutes to an approved rubric. One role is enough to see what it finds.
Nothing is read until you approve the rubric.