Showcase · Vol. 01
Fine-tune reports, built in one prompt.
Reference builds from the recipe library. Each report is a static TanStack page rendering a JSON artifact produced by a Tinker training run in Lovable's Linux build sandbox — no runtime API calls.
Vol. 01 · No. 01 · SFT-InstructKrump Callout LoRA — fine-tune Qwen on session callouts.
A single Tinker SFT run on 80 authentic krump session callouts. Loss curve rendered from the training artifact, before/after samples show the base model's generic reply next to the tuned LoRA's on-house one. Built in one Lovable prompt to prove the 5-credit budget is real.
Open the report →Vol. 01 · in production
More reports land here as builders ship them.
Pick any recipe from the index, paste its mega-prompt into Lovable with your TINKER_API_KEY set, and your build will join the showcase.