We plan to run a competition with a prize pool as an experimental way of finding amazing candidates. Has anyone seen this work out?

For the top submissions we will extend an offer to join the team.

For the winning submission, we also plan to ship their solution into Danswer so that their work has impact beyond the competition. The goal will be to create a solution for query expansion to use in RAG. To be able to actually productionize it and to encourage more creative thinking (can’t just finetune a pretrained encoder decoder transformer), we have a few requirements:

[1] If it’s deep learning based, it must be capable of running on CPU (up to 50 tokens in 200ms or less on a 4 cpu core machine) [2] It must not rely on any proprietary or restrictively licensed software

I know there is Kaggle but I don’t believe they allow constraints like the above ^

What is the best way to run a competition like this?