I’m in the process of building a service that will allow contributors to offload model computation to the cloud.
The MVP feature set is going to be very basic, consisting of an API to upload the model Python code to and an email response containing presigned URLs to the generated artifacts. These would likely be the tournament and validation prediction output CSVs.
Future features could include directly invoking the Numerai validation API, in-line code editing, improve cost performance, and payments based on model performance.
I’m participating in the #BuildSell30 challenge on Twitter so the goal is to have the MVP done by the end of the month.
Currently, most of the backend is complete and I’ll be tweaking and fixing infrastructure through the weekend. After that’s done, I’ll pivot over and build a small wrapper UI or CLI and invite beta users - if you’re interested, sign up for email updates here.
Best Case Outcome:
Pretty straightforward in that a decent number (I don’t have a specific number) users would be using the platform and that we see an increase in submitted models as a result.
Worst Case Outcome:
Low user engagement in which case I might open source the code and take down any running instances in my own account.
This would be hitting the MVP of exposing a functional API and likely wrapping it with a UI or CLI depending on what the community is looking for.
I don’t really need payment but I’ll track the hours regardless. It’s an evenings and weekends project so probably something like ~10hr / weekday evenings and ~10hr / weekend.
Let me know what you all think regarding feasibiltiy, features, and gathering more feedback from the community at large. I’m definitely interested in understanding what features and interface the community would prefer for something like this. If you’re interested in collaborating (project stack is Python and AWS serverless offerings) DM me!
P.S. Thanks to @aventurine for bringing me in and recommending that I submit a proposal!