Blog
Notes from building ModDog: running AI moderation on self-hosted GPUs, and what we learn the hard way.
Good models come in small packages
We assumed our moderation model kept missing things because it was too small, but the bigger model we trained failed the same way. Fixing the data fixed both, and the small one is what runs today.
AMD and Intel are dipping their toes into AI at your expense
Intel archived its own AI software while I owned the card. AMD shipped a flagship model its own "AI PRO" card can't run. If they won't support these cards, they owe us the firmware.
Six weeks with Intel's Arc Pro B70 in a production inference fleet
A $1,000 32GB workstation GPU that crashes under sustained LLM serving load. Everything we ruled out, every workaround that failed, and what to buy instead.