Skip to content
boson
How it works

One pipeline, from invite to evidence.

Compose the rounds that predict performance for the role, invite candidates with one link, and review a shortlist built from scores you can inspect.

  1. 01

    Compose the pipeline

    Pick the modules that fit the role and put them in order. Versioning keeps a candidate on the pipeline they started.

    A pipeline is an ordered list of modules, each with its own config, deadline and required flag. Versioning is intentional: editing a pipeline does not change an application already in flight.

    • Choose from the module registry
    • Set per-step deadlines and required flags
    • Save a new pipeline version without disturbing in-flight candidates
  2. 02

    Invite with one link

    Send an individual single-use link or a public link with an expiry and a use cap.

    Links are rate-limited and tokenised. Candidates sign in before taking a round, so every submission has an identity and a consent record attached to it.

    • Individual links prefill the candidate email
    • Public links can expire or close after a set number of uses
    • No captcha: identity comes from sign-in
  3. 03

    Candidates sign in and take rounds

    One portal, one clear next step, a visible deadline and autosave throughout.

    The candidate sees the consent gate and the next round with its deadline. Drafts save automatically and reconnect handling keeps a dropped connection from ending the round.

    • Plain-language status instead of internal stage names
    • Autosave and reconnect handling during a round
    • Separate consent for AI evaluation and each recording type
  4. 04

    Grade with evidence

    Deterministic parts stay deterministic. LLM grading returns structured scores, reasoning, confidence and evidence.

    Multiple-choice keys and SQL results are scored deterministically today. Free text is judged with structured output, multiple samples and a confidence measure. Every score stores its prompt and model version.

    • Verbatim evidence for resume quotes
    • Median and spread across judging samples
    • Low-confidence runs flagged for human review
  5. 05

    Review the ranked shortlist

    Open a candidate to see the dimension breakdown, evidence, confidence and any override history.

    Ranking uses per-job weights and shows how much of the result is stub or low-confidence. Recruiters can override with a required reason; the audit trail records who changed what and why.

    • Per-dimension breakdown with reasoning
    • Override with a required reason
    • Replay when a round records one

One registry, eight module types.

The registry defines eight module types. Each available module ships its own config schema, candidate experience, runner and grader, and new modules plug into the same pipeline instead of changing the engine.

Resume review

Available today

Pasted resume text, graded against the job with verbatim quote evidence.

Questionnaire

Available today

Written questions in the portal. Deterministic checks for choices and required fields, LLM grading for free text.

Assessment

Available today

Multiple-choice and SQL questions with deterministic scoring and a read-only SQL preview.

Coding

Coming soon

A timed problem in a sandbox with test-case scoring, editor replay and process grading. Needs Judge0 before it can run.

System design

Coming soon

An open-ended design answer with a text and/or diagram deliverable. The grader is still a stub.

Voice screen

Coming soon

A spoken screening round with audio and transcript, plus optional webcam off by default. The voice engine is not built yet.

Async voice Q&A

Coming soon

Recorded spoken answers to a fixed prompt list, taken at the candidate's own pace. The voice engine is not built yet.

Human interview

Coming soon

A live or offline interviewer step scored from a structured scorecard. The candidate and recruiter flow is still being built.

Two views of the same pipeline.

The recruiter workspace and the candidate portal read the same state. Nobody has to guess what happened between rounds.

For candidates

  • One clear next step and deadline
  • Autosave and reconnect handling
  • Plain-language status, not internal stages
  • Consent before any evaluation or recording
  • Results visibility controlled by the recruiter

For recruiters

  • Per-job weights and a ranked breakdown
  • Evidence and confidence on every score
  • Override with a required reason
  • Audit trail on scores and access
  • Replay when a round records one

See the shortlist before you commit to a process.

Start with a single role and the resume, questionnaire or assessment round. The same pipeline can grow into coding, system design, voice and human interview as those modules ship.

Get startedRead the product notes