pathwize

How it works

Every project starts with a test on your own items. Only experts you approve do paid work, and they keep being checked while they do.

Step by step

Six steps from brief to delivery.

A typical pilot takes about three days from your brief to approved experts. After that, the team keeps working on your full data set.

  1. 01

    Brief and examples

    You share sample data, your classes or rubric, and at least one example per class plus rules for borderline cases. We turn it into the briefing every expert reads.

    YouDay 0
  2. 02

    Expert matching

    We open the task to the pool for your field, for example civil engineers for construction waste or lawyers for legal evaluation. Experts accept your project terms before they see data.

    PathwizeDay 0 to 1
  3. 03

    Test on your items

    Each expert works through the same test items, usually 50. Scores are calculated against your reference answers, or you grade the answers yourself.

    ExpertDay 1 to 3
  4. 04

    You approve the team

    Compare scores, speed and individual answers in the portal and approve the experts you want. Experts who failed can practice and retry.

    YouDay 3
  5. 05

    Production with checks

    Approved experts label or rate your data. Reference items, second reviews and the review queue run in the background.

    ExpertOngoing
  6. 06

    Delivery

    Download CSV, JSON or COCO from the portal, or receive results through the REST API and webhooks as they come in.

    PathwizeContinuous

Before you start

What we need from you.

  • A sample of your data (images, documents, text or prompt and answer pairs)
  • Your classes, labels or grading rubric
  • At least one example per class, ideally a few
  • Rules for borderline cases and what to do when unsure
  • Reference answers for the test items, if you want automatic scoring
  • Your expected volume and deadline after the pilot

Your decision

You see every score before anyone is paid.

In the portal you compare each expert's test score, speed, cost per label and individual answers. Approve the ones you trust, reject the rest. Experts who did not pass can review their mistakes, practice and try again.

  • Score per expert against your reference answers
  • Time per item and cost per label from the test
  • All test answers downloadable as CSV
  • Approve or reject with one click, the expert is notified

Two ways to work

Managed or self-service.

Most teams start here

Managed

We set up the task, the briefing and the test with you, run recruiting and keep an eye on quality. You approve experts and receive data.

For repeat projects

Self-service

Create a business account, set up tasks and tests in the portal or through the API, and invite expert pools yourself.

FAQ

Questions about the process.

How do I start?

With a free pilot: send up to 50 items, your guidelines and an example per class. Several experts take a test on them, you see their results and choose who continues. Results usually arrive within 72 hours.

Can I run it myself?

Yes. Create a business account, set up a task in the portal or through the API, and invite expert pools. Or let us set it up with you.

Which data types and formats?

Images, documents and text, plus prompt and response pairs for language models. Results export as CSV, JSON or COCO, or through the REST API and webhooks.

Which languages?

English and German are the most common. Other European languages depend on the pool for your domain, so ask before you start.

Ready when you are.Start with 50 items.

The pilot is free and comes with no commitment.

Request a free pilot