JevKit – Local fine-tuning and eval studio for JEV models

Fine-tune small classifiers locally with little data, no cloud required

Problem

Developers are curious about fine-tuning small LLMs but lack guidance; JEV models gain traction for efficiency, yet easy local tuning and eval tools are missing. Projects like Shrewd and JevEval show demand but no complete workflow.

Solution

A desktop app or CLI that guides users through dataset import, model selection (JEV-compatible), one-click fine-tuning, and built-in eval (accuracy, confusion matrix). Export models for local inference.

Users

ML engineers, data scientists, indie devs building specialized classifiers on-device.

MVP

Import CSV/JSONL; integrate one or two small models; basic metrics; export to ONNX or PyTorch.

Revenue

Free community + paid pro (more models, hyperparameter search, team collab), one-time or subscription.

Why now

Edge AI and privacy push local fine-tuning; JEV and similar models are hot, but tooling is nascent—indie devs can define the standard.