Read and match
Everyone shows you a number. The question is whether it opens.
A score you cannot take apart is a score you cannot defend, and defending it is the entire job — the moment a client asks why their favourite CV was passed over, a confident number is worth nothing and a line from the CV is worth everything.
What comes back
One row per requirement, and the line that answers it.
The role's requirements are established first — from the description and, where it was thin, from the conversation that filled the gaps. Then each one is answered separately: shown by direct evidence, inferred from context, or evidenced by nothing at all.
The requirement nothing answers is stated, not averaged away. That is the row a client will ask about, and the one a summary figure is designed to smooth over.
- Streaming pipelines in productionShown“Owned the event pipeline handling ~40k messages/min across three regions.”Experience · 2022–2024
- Data modelling for analyticsShown“Redesigned the warehouse schema; cut nightly build from 6h to 90m.”Experience · 2021
- Leading a small teamInferred“Ran the on-call rota and onboarded two engineers.”Experience · 2023 — not stated as a reporting line
- Regulated-industry exposureNothingNothing in the document addresses this. Not inferred, not scored down — left open, because it is the question to ask in the interview.
Before you spend anything
The first pass costs nothing and still names the blocker.
Most evaluations of AI matching stall on the same objection: I am not running an AI call across nine thousand records to find out you have nothing. Reasonable. So the first pass does not make one.
One action reads every candidate you hold against the role and sorts them into four bands — and for everybody it set aside, it names the requirement that did it. Hard blockers are kept separate from preferences you are allowed to override, because not a fit and not a fit on the notice period you invented are different answers.
How they sorted
- Strong12Every stated requirement answered.
- Promising47One preference short, nothing mandatory.
- Stretch96A real gap, and a reason to look anyway.
- Hard gap258Something mandatory is missing.
What set them aside
- Work authorisationHard104
- Mandatory: production KafkaHard87
- Notice period ≤ 30 daysOverride41
- Preferred: on-site or hybridOverride19
- Above the budget ceilingOverride7
The overridable rows are the interesting ones. Forty-one people are out on notice period alone — a number worth knowing before you decide the notice period is non-negotiable.
- It runs on what you already hold —
No AI call, no per-record cost, and no waiting. It is the pass you run before you decide whether a deeper read is worth it.
- The deeper read is a choice, not a default —
The evidence-cited verdict is a separate entitlement, off unless it is turned on for your workspace. Nobody discovers it on an invoice.
Beyond keyword overlap
Career shape, and skills read in context.
Two candidates can share every keyword and have had completely different careers. A specialist in an adjacent discipline writes the same vocabulary as the person who actually did the job; ten years owning a small workstream reads identically to ten years contributing to a large programme if you are counting tokens.
So the read includes the shape of the career — the trajectory, whether the narrative in a summary is supported by the evidence in the projects — and whether a skill appears as something done or something listed.
- Gaps and highlights, stated —
What the role asked for that this person does not demonstrate, and what stands out positively.
- Defensible line by line —
A shortlist you can walk a client through is a different object from a ranking you have to defend with conviction.
- Your corrections become the truth —
Edit a field the reader filled and the correction is captured against that record, and the corrected value is what the next round treats as true. It is a loop inside your workspace, not a model somebody else trains.
- Names off, if you want them off —
One switch hides names, email addresses and current employers across every result. It persists for you, so a review that is meant to be blind stays blind after the tab is closed.
Objections
What a careful buyer asks here.
›Everyone says they do AI matching. What is actually different?
Two things you can check on the call. The first pass is free, instant and names the blocking requirement per person. The deeper pass returns a row per requirement with the source line attached, including the requirements nothing evidences. We do not claim AI matching as a category — we claim this particular output.
›There is still a number at the top. Why?
Because a sorted list is useful and pretending otherwise would be precious. The difference is what happens when you press it: it opens into the requirements that produced it, and the ones nothing answered are listed rather than absorbed.
›Will it reject people automatically?
No. It does not advance or reject anyone. Where it cannot support a claim with a line from the source, it drops the claim rather than softening it.
›Can we stop it seeing names and employers?
Yes — one switch, and it holds across searches and results until you turn it off.
›Does it work for non-technology roles?
Yes — the requirements come from the document, not from a taxonomy of job titles. Bring a non-technology role to the demo and we will run it.
›What happens when it gets a field wrong?
You correct it on the record. The correction is captured and the corrected value is what the next extraction round reads as truth, for your workspace only.
Book a demo
Run the free pass on the call.
Bring a role and your database. The first thing worth looking at is not the shortlist — it is the list of people who were excluded on a preference you could have overridden.