Hey everyone ๐
I'm building something I wish existed when I was debugging training runs โ an AI agent that automatically detects and explains anomalies in LLM/ML training.
Things like:
โข Gradient vanishing in specific layers
โข Attention head collapse
โข MLP activation drift
...and instead of just flagging a number, it traces the root cause and suggests fixes. Like having a senior ML engineer watching your run 24/7.
I'm still in early validation โ no product yet, just a working prototype and a lot of conviction.
Looking for 10 ML engineers to give me 15 minutes of honest feedback. In exchange: free lifetime access when it launches + your name in the credits.
If debugging training runs has ever cost you hours (or days), I'd love to talk. Drop a comment or DM me directly.