How Bloom scores — and what it will never do
Bloom is an AI recruitment engine built around the hiring manager's judgment. This page says, in plain words, what the engine does with an application, who is in charge of it, and the things it is built never to do.
The hiring manager owns the bar
Before anyone is scored for a role, Bloom drafts a criteria card and the hiring manager confirms it: the pass-or-fail deal-breakers, then the criteria that matter most and the nice-to-haves, each with a weight the hiring manager can set. Bloom scores against exactly that card — not against a generic idea of a good candidate.
Cards are versioned. A change creates a new version, and every score records which version it was scored against, so nothing is silently rewritten underneath an existing score.
Bloom recommends. It never decides.
Every placement Bloom makes is a recommendation to the hiring manager. Nothing advances and nothing is rejected without the hiring manager's say-so. The engine's job is to make a large pool reviewable; the decision stays with the person accountable for it.
Every score shows its work
Every score opens to the criteria it was scored against, the weight each one carried, and the evidence in the application that supports it. A hiring manager can see why a candidate landed where they did, and because the placement is a recommendation, they can overrule it.
A name is not a criterion
Bloom scores against the hiring manager's criteria card, read for the role it was written for, and nothing else. The candidate's name, email, phone number and LinkedIn link are removed from the résumé before it is scored, so a name is not a criterion. Every score opens to the criteria and the evidence that produced it, so what moved a score is visible.
Read before scored
Bloom reads the whole résumé and every pre-screening answer before it scores anything. Scores are built from what the candidate actually wrote, not from keyword matches against a job title.
Uncertainty alone never archives
When the engine cannot tell whether a deal-breaker is met, it does not guess, and that alone never archives the candidate. The candidate is surfaced to the hiring manager with the open question attached. Only candidates who are clearly out — a missed deal-breaker or a clear mismatch on the criteria — are archived, and the reason is recorded with them.
Four states, and what each one means
- Of Interest — cleared the hiring manager's bar. Reviewed first.
- Consider — close, with one specific reason attached. Lifted out of the pool so nobody misses them.
- Pool — scored and explained, kept on file. Nothing here is lost.
- Archived — clearly out, with the reason recorded.
Nothing advances without the hiring manager
Bloom does not advance a candidate, send a decision to a candidate, or close a role on its own judgment. It recommends and it drafts; people make the hiring decisions. Bloom never moves a candidate from one stage to the next.
Where the data lives
Bloom's database runs in Supabase's Canada Central region, and the application is served from Vercel's Montréal region. Bloom's database, which holds candidate records, is in Canada. The Privacy Policy and the candidate notice describe what is collected and for how long.
What Bloom will never do
- Decide who is hired. Bloom recommends; the hiring manager decides.
- Charge per seat. Every plan is one flat monthly price with unlimited users.
- Hide how a score was reached. Every score opens to its criteria, its weights and its evidence.
- Archive a candidate because a deal-breaker is uncertain.