The sorting is done before you open your inbox
For recruiters drowning in CVs.
On every application, a French AI reads the CV, scores it out of 100 on public criteria and summarises the profile in two sentences. You start with the best.
30 days free, no card.
Ranked automatically
Applications arrive sorted by score, with relevant profiles rising to the top of the pile.
Public criteria
Skills 35%, experience 35%, education 15%, motivation 15% — the weight of each criterion is shown.
Two-sentence summary
A profile summary, up to 3 strengths and 3 concerns: you understand the score at a glance.
Human oversight
The AI suggests, you decide. No automatic decisions — AI Act compliant.
Key capabilities
- Score out of 100 computed automatically on every application
- 4 public weighted criteria: skills, experience, education, motivation
- Two-sentence summary + up to 3 strengths and 3 concerns
- Manual re-scoring from the application page
- Automatic retry of failed scorings (cron)
- French/European model (Mistral), data hosted in France
A score you can explain
The global score is recomputed server-side from the contractual weights (35/35/15/15) — never simply trusted from the model. You can re-run a scoring by hand from the application page.
Criteria and their weighting are public and identical for every candidate on a given job: a recruiter, like a candidate, can understand where a score comes from.
Built to be fair
The analysis ignores what should not matter (age, gender, photo, address), and an evaluation harness regularly checks that two identical CVs get the same score regardless of the name.
An attempt to manipulate the CV to inflate the score (hidden text, injected instructions) is detected and flagged, never rewarded.
Robust and traceable
Temperature 0, a hard timeout per call, one immediate retry on invalid JSON before marking a scoring as failed: the score is stable from run to run.
Tokens, duration and retries are logged. An AI usage card (scorings, tokens, average duration) is available on the admin side for monitoring.
Frequently asked questions
Can the AI reject a candidate on its own?
No. The AI proposes a score and a summary; the decision always stays with the recruiter (human oversight, AI Act).
Which model powers the scoring?
A French/European model (Mistral) via Scaleway Generative APIs, hosted in France. No candidate data leaves the EU.
Is the score the same every time?
Yes, within a few points: temperature is 0 and the global score is recomputed from fixed weights, which guarantees high stability (σ ≤ 5 on our test set).