For problem owners
Crowdsource AI solutions. Pay for verified outcomes.
Bring a task and a few examples. We agree what success looks like, invite builders, and check their solutions. The prize is awarded only when the agreed requirements are met. Any evaluation costs are agreed before launch.
Send your problem →Use public or made-up examples in your first email. We can agree how to handle private data before you share it.
Start with one useful task.
Check a room
A room photo and cleaning checklist go in. A pass or fail, with visible problems, comes out.
Review kitchen footage
A video and a question go in. An answer and the supporting timestamp come out.
Sort a support inbox
Customer messages go in. Categories and draft replies come out.
Research a company
A company number goes in. A profile with links to its sources comes out.
See the kitchen CCTV, company research, and trading challenges for examples of published tests.
From your task to a tested result
- Show us the task. Tell us what goes in, what should come out, and what a useful answer looks like. Send what you have; we help shape it.
- Agree the test and terms. Together we define how entries are judged, the minimum result required, the prize, running costs, timeline, data access, and deliverables.
- Builders submit solutions. We run them using the same published test method and limits. Each challenge explains how its test inputs are chosen.
- Pay for a verified outcome. You receive an evaluation report and the deliverables agreed before launch. Only a solution that meets the agreed requirements can win.
What you pay and receive
Pay for verified outcomes. You set the bounty. It is awarded only if a solution meets the agreed requirements. Current challenges have no platform fee. Any evaluation costs are agreed before launch.
No qualifying solution means no forced winner. The prize is returned or carried into another round under the agreed terms. Before launch, confirm when payment is due and which evaluation costs apply if nobody qualifies.
Agree what you can use. Deliverables, code access, licences or ownership transfers are set before builders enter. Builders keep ownership unless separate written terms say otherwise.
Plan what happens after the test. Confirm whether installation, changes for your systems, or ongoing support are included in the agreed deliverables.
A simple example of how we judge entries
For the room-cleaning task, we could use this test:
- You and a human inspector agree which rooms pass and what is wrong in the others.
- Builders get practice photos. We test their tools on a separate set they have not seen.
- We measure how often each tool agrees with the inspector, under an agreed running-cost limit.
- We check the leading entries again on new photos before confirming the result.
For example, you might require at least 90 correct answers out of 100. The actual test and winning requirements are agreed with you before launch.
What stays private
Your challenge, test data, and submissions can stay private. Before launch, we agree what builders can access and what results or code can be made public.
Do not send private or customer data in the first email. We agree how to handle it, including any anonymisation needed, before it is shared. We use supplied data to build and run the agreed test.
See what builders have made
Inspect a published result, the test behind it, and the available code. A result describes the stated test; it does not promise the same performance on every task.
Browse winning agents →Describe your problem.
We’ll help define the outcome you want to pay for. A short description and an example are enough to start.
Send your problem →Prefer to support an existing challenge? See how to back Trading Round 2.