I applied through an employee referral. The process took 2 weeks.

Was given an 8-11 hour take-home test where you are required to use AI to fully complete the project in one work day. I had to build a fully deployed, production-ready text-to-speech server using open-source models that needs to handle heavy concurrent loads.

I built this server, deployed to GCP Cloud Run using Terraform, including authentication (user API keys), and I also added the ability to proxy paid API TTS services like Gemini TTS, OpenAI TTS, ElevenLabs TTS. I then built an admin console to control this (not part of the original spec but really tied the project together).

The admin console used OAuth to secure access, had a logging interface for searching logs, and had a metrics dashboard for detailed metrics about what models are being used. Everything looked polished, and I submitted it after about 9 hours of intense work, less than 24 hours after they gave me the test.

There were no leaked API keys or any other obvious reasons that this project "wouldn't meet the bar". I had to record a video describing the project, but they asked for the video to be about 5 minutes, so you have to assume they would give you time to explain things at some point. Nope.

I was feeling confident about my decisions and looking forward to discussing the reasons why I made them... But Luma replied that my project did not meet their bar. I asked for direct feedback as to why, multiple times, but they avoided the question and then didn't reply.

I looked at my service logs. They never even tried to access the speech server. It makes me think that they never even evaluated my work. That is pretty soul-crushing after putting a massive amount of work and $70 of my money into token costs.