Fine-Grained Human Feedback for LLMs

A full-stack web application to collect fine-grained human feedback on LLM-generated responses.

Tech Stack

• Frontend: React (Vite), Tailwind CSS
• Backend: Flask (Python), REST API
• Database: Supabase (PostgreSQL)
• AI Integration: Google Gemini API

Features

1. Enter a prompt and generate a response from an LLM.
2. Select any part of the text (span) to annotate.
3. Provide multi-dimensional feedback (helpfulness, accuracy, clarity, safety).
4. Mark spans as good/bad and rewrite bad spans.
5. Generate an improved response based on the aggregated feedback.

Showcase