Spaces:
Sleeping
A newer version of the Gradio SDK is available:
5.38.0
title: ReadRight
emoji: 📚
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.33.0
app_file: app.py
pinned: true
license: mit
short_description: AI-powered reading coach that adapts to help you learn!
tag: agent-demo-track
🦉 ReadRight - AI Reading Coach for Kids
Overview
ReadRight is a simple app designed to help students, especially in non-native English speaking countries, improve their English reading skills. Many students don’t have access to personal tutors or extra help with pronunciation and reading fluency. This app uses AI to fill that gap, making it easier for anyone to practice reading out loud and get helpful feedback.
With ReadRight, students can read short, age-appropriate stories generated just for them. The app listens as they read, checks their accuracy, and gives friendly, easy-to-understand feedback. If a student struggles with certain words, the app will help them practice those words and even create new stories to reinforce learning.
This project is an MVP (Minimum Viable Product) submitted for Track 3: Agentic Demo Showcase at the hackathon, showcasing the power of AI agents through a multi-component architecture built with Gradio, smolagents, and advanced AI services.
🎯 Motivation
English pronunciation can be tough for non-native speakers—words like "knight," "through," or "psychology" don’t sound how they look. Many students feel shy about speaking because they lack a safe, affordable space to practice without judgment. ReadRight tackles this by offering personalized, AI-powered reading practice that adapts to each student’s level and interests, helping them gain confidence and fluency.
🛠️ How It Works
ReadRight uses a modular setup with Gradio Spaces API and smolagents to deliver a dynamic, adaptive learning experience. It relies on AI agents in two key phases, with the LLM driving the flow in a multi-step agentic workflow, especially when adapting to student progress.
Story Creation and Audio:
- Students enter their name, grade, and a topic they like through a simple Gradio interface.
- Story Generation Agent: Powered by LLM(can be any) and smolagents’ tool-calling features, this agent crafts engaging, personalized stories tailored to the student’s grade and interests. It automatically adjusts story length, vocabulary, and complexity—for example, using simple words for younger kids or richer sentences for older ones—making smart choices about content without needing extra user input.
- Audio Creation: Hugging Face TTS (NihalGazi/Text-To-Speech-Unlimited) turns stories into natural-sounding audio to guide pronunciation.
This phase starts with user input, but the story agent works autonomously to shape the content.
Feedback and Adaptive Learning:
- Once a student records their reading, a multi-step agentic workflow kicks in, powered by smolagents:
- Speech Recognition: Whisper Large V2 (abidlabs/whisper-large-v2) transcribes the student’s reading accurately.
- Text Comparison: A custom Python engine using
difflibcompares the transcription to the original story, spotting errors and mispronunciations. - Feedback Agent: The LLM generates friendly, detailed feedback with pronunciation tips tailored to the student’s performance. It also decides when to create new stories that focus on words the student found tricky, forming a loop where it adapts content based on performance without needing extra user prompts.
- This phase shines as a multi-step agent, with the LLM analyzing data and choosing next steps to create a personalized learning path.
- Once a student records their reading, a multi-step agentic workflow kicks in, powered by smolagents:
Together, these phases form a semi-autonomous system: story creation starts with user input but uses an agent to craft tailored content, while the feedback phase is highly agentic, dynamically adjusting to each student’s needs.