Start free

End to end

From a client requirement to somebody’s first day.

Fifteen stages, one set of records. Every step below is a real surface in the product, and every AI step has a human review in front of the business action it leads to.

The flow

What happens, in order.

  1. Client requirement becomes a job

    The role is opened with its own hiring stages. Source

  2. Candidates enter the system

    Career page, bulk upload, referral, or added by hand. Source

  3. Resumes are parsed into profiles

    Structured fields extracted from whatever format arrived. Understand

  4. A human reviews the extraction

    The review diff shows exactly what would be written. Understand

  5. Candidates are matched to the role

    Deterministic requirement checks, then semantic matching. Understand

  6. An AI agent runs the screening call

    With a consent disclosure and a hard attempt cap. Screen

  7. The call becomes a structured report

    Readable answers, not an hour of audio. Screen

  8. The recruiter reviews and decides

    AI output is an input to a person, never an action. Screen

  9. The application moves through the pipeline

    With SLA tracking on every stage. Decide

  10. Interviews are scheduled with a brief

    Optionally including a live coding round. Decide

  11. Evaluations and feedback are recorded

    Against the application, in a comparable shape. Decide

  12. The client reviews the shortlist

    And their feedback turnaround is measured. Close

  13. Offer, then hire

    The document checklist opens itself on hire. Close

  14. Onboarding documents are collected

    With an overdue sweep before the start date. Close

  15. Everything is measured and automated

    Analytics over the funnel, automation for the repetition. Operate

By role

The same system, four different jobs.

A hiring manager does not want a recruiting tool, and a candidate never signed up for one. Both still have to be served.

Agency recruiter

Many clients, many roles, and submissions to track.

  1. Take the requirement
  2. Search candidates already on file
  3. Bulk-add new resumes
  4. Screen and shortlist
  5. Submit to the client
  6. Track feedback turnaround

In-house talent team

Fewer roles, deeper process, more stakeholders.

  1. Open the role with its stages
  2. Route inbound applications
  3. Match and screen
  4. Coordinate interview panels
  5. Record structured evaluations
  6. Hand off to onboarding

Hiring manager

You do not want a recruiting tool. You want a decision.

  1. See only your roles
  2. Read the shortlist with reasons
  3. Take the interview with a brief
  4. Leave structured feedback
  5. Approve the offer

Candidate

They never signed up for this product, and never have to.

  1. Apply from a public link
  2. Take a screening call with a disclosure
  3. Open the coding round from a link — no install, no login
  4. Receive interview details
  5. Upload documents on hire

The AI safety model

Where the model stops.

The pipeline runs in one direction and never skips a step: raw data, then AI, then a structured output, then validation, then human review, and only then a business action.

Raw data → AI → structured output → validation → human review → business action

AI proposes. A person decides.

The pipeline is fixed and it runs in one direction: raw data, then AI, then a structured output, then validation, then human review, and only then a business action. AI output never writes straight to a trusted table.

  • Every AI function has a named purpose and a validated output shape.
  • Generated summaries are checked in code against the real figures.
  • When the AI provider is down, the manual workflow still works.

Anything that contacts a person fails closed.

Calls, emails and messages are disconnected by default. Connecting one takes an explicit action, sending needs confirmation in the UI, and there is a hard cap on attempts.

  • A consent disclosure that cannot be disabled opens every call.
  • Opt-out links work without a login, because candidates have no account.
  • Recording someone without telling them is unlawful in many places, so the product is built so it cannot happen.

Every organization's data is isolated in the database.

Each tenant's rows are separated by row-level security in PostgreSQL. The tenant is resolved from the session — never from a request body, a query parameter or a header.

  • Isolation holds even for a caller talking to the database directly.
  • Provider credentials are encrypted and readable only through a privileged path.
  • An append-only audit log records who changed what.

Open one role and see.

The manual workflow works on its own. The AI is assistance you switch on, not a dependency you inherit.

Create an accountBack to overview