Structured extraction
CVs and job ads are extracted into defined fields rather than summarised into prose, which is what makes the output checkable at all.
All features
AI Orchestration & Compliance
AI you can audit, not AI you have to trust.
The question a search firm should ask about AI is not whether it is impressive but whether its output can be defended six months later. RayCruit treats AI as a governed capability: every output checked, every call logged, every downstream artefact aware of when its source changed.
- Every output checked against a fixed structure before it is shown
- Versioned prompts with provider and model recorded
- Automatic staleness flags on dependent artefacts
- Per-action routing across four providers without redeploy
The problem
Ungoverned AI creates work that cannot be defended.
A recruiter pasting a CV into a chat window gets a useful summary and creates four problems: nobody knows which model produced it, nobody can reproduce it, it silently overwrote the recruiter's own judgement, and when the CV is updated the summary stays wrong. In a regulated, client-facing business, that is not a productivity gain.
- No record of which model, prompt or input produced a client-facing claim
- Free-text output cannot be validated, so errors reach the client
- AI output overwrites human assessment without anyone noticing
- Artefacts go stale silently when the underlying CV or mandate changes
Every AI interaction structured, validated, logged and reversible.
AI in RayCruit is confined to well-defined actions with defined inputs and outputs in a fixed structure. That constraint is what makes it auditable.
Every output is checked before you see it
Every AI output is checked against a fixed structure before it reaches a recruiter's screen. Anything that does not fit is not displayed.
Versioned prompts
Prompts are versioned artefacts. Any past output can be traced to the exact prompt version that produced it.
Provider and model logging
Every call records its provider, model, prompt version, input and output — and can be streamed live so administrators can watch the system work.
Staleness detection
Change a CV or a mandate and every dependent analysis, matrix and export flags itself for regeneration. No client receives a summary built on an outdated source.
No silent overwrite
AI fills blanks and proposes changes. It does not overwrite what a human wrote, and every suggestion is visibly a suggestion until accepted.
Human review by design
AI output is decision-support material. The workflow requires a person between generation and anything client-facing.
Multi-provider routing
RayCruit supports OpenAI, Anthropic, Google and Mistral, and any individual AI action can be routed to a different provider without a redeploy. The managed RayCruit service currently processes with Mistral AI in Europe — see the Privacy Policy for the current subprocessors.
Rules where it matters
The core match score is computed, not generated. AI is used for judgement-shaped tasks — risk, positioning, interview preparation — not for the number the shortlist is ranked on.
The lifecycle of one AI action
Every AI action in the platform follows the same governed path.
- 01
Defined trigger
A recruiter runs a specific action against specific structured input.
- 02
Versioned prompt
The action uses a versioned prompt routed to the configured provider and model.
- 03
Structure check
The response is checked against the expected structure before it is accepted.
- 04
Review
The output is presented as a suggestion for a person to accept, edit or discard.
- 05
Dependency tracking
If the source later changes, everything derived from it is flagged for regeneration.
How search firms use it
Client data-protection review
An enterprise client asking which models process their candidate data gets a specific, configurable answer.
Model policy
A firm that requires a particular provider for certain data routes those actions accordingly, without changing anything else.
Post-hoc investigation
When a client questions a claim in a submission, the logged prompt version and inputs make the answer reconstructable.
What changes
AI that passes review
Governance answers the questions procurement and legal actually ask.
No stale client material
Dependency tracking prevents the most damaging category of error.
Recruiter authority preserved
Human judgement stays authoritative by design, not by convention.
See it in motion
The platform, end to end, in two minutes.
How the modules hand over to each other in the real product — customer and mandate, candidate and matrix, the workspace where a pairing is decided, and the material that goes to the client.
2 min
Continue across the platform
Placement work does not stop at one module. These parts of RayCruit connect directly to this one.
Matching & Matrices
Rule-based scoring across 12 dimensions and 90 criteria, with evidence on every line.
Learn moreGovernance, Permissions & Audit
Permission groups, confidential mandates, enterprise sign-in and an append-only audit trail.
Learn moreCandidate Intelligence
Structured CVs, skills, languages, eligibility and a reusable candidate matrix.
Learn moreSee the modules working together.
A demo follows one mandate across every module, which is the only way the connections become obvious.