Rebiha: Fine-tune LLMs with ready-made datasets, no infrastructure

Hey PH 👋 Malek here, solo founder of Rebiha. I've been doing AI fine-tuning work for a while, and the actual model training was never the hard part. The hard part was everything around it — provisioning a GPU, keeping Docker containers from silently dying, watching a job idle and burn money because I forgot to shut the instance down. None of that has anything to do with whether your model turns out good. So I built Rebiha to make that whole layer disappear. Pick a base model, pick from 35 ready-made datasets (or bring your own), configure a few hyperparameters, and hit train. The GPU spins up when the job starts and shuts down automatically when it's done. You pay per job, not for idle compute you forgot about. I'm building this solo — backend, frontend, infra, all of it — so I'm around in the comments today to answer anything: model selection, dataset quality, pricing, what's on the roadmap, whatever. If something's confusing or missing, tell me straight, I'd rather know now.