PromptLab
2026 • CS 568 Group Project at the University of Illinois Urbana-Champaign
Description
A web app that scores and rewrites AI prompts, powered by a fine-tuned local model. In Check Prompt mode, users paste a prompt and get scores across clarity, specificity, tone, and ambiguity, plus a rewritten, optimized version. In Generate Prompt mode, users describe an idea in plain language and get a polished, ready-to-use prompt.
Tech Stack
React 18 + Vite frontend. Flask backend serving a fine-tuned SmolLM2-360M-Instruct model with custom LoRA adapters: one for BPO-style prompt optimization, one trained on HelpSteer for scoring. Google Gemini generates the auxiliary explanation text.
