Inspiration
We wanted to build something that could make studying easier for school students, especially when they struggle with difficult concepts or are more comfortable asking questions in Hindi or Hinglish. We were inspired by the idea of having an AI teaching assistant that students can talk to, ask questions to, and learn from visually instead of relying only on textbooks or text-based chatbots.
What it does
VidyaSaarthi is an AI-powered learning assistant for school students. Students can type or speak a question and get a simple explanation based on their class level. It supports English, Hindi, and Hinglish, generates educational visuals to explain concepts, and can create quizzes to help students check their understanding. It also provides text-to-speech, making the interaction more natural and useful in a classroom setting.
How we built it
We built the application using Python and Streamlit. We used Groq-hosted LLMs for generating explanations and quizzes, Whisper for converting voice input into text, and Edge TTS for voice responses. For visual learning, we experimented with Mermaid diagrams, Wikimedia images, and AI-generated visual content. We kept the project modular by separating the AI, speech, prompt, quiz, and visual-generation components.
Challenges we ran into
A major challenge was making the voice interaction work reliably, particularly with Hinglish and mixed-language speech. We also faced API rate limits and model deprecations while developing with LLM APIs. Getting the LLM to consistently return structured output for our visual and quiz components was another challenge. We had to add validation, fallbacks, and modify our prompts several times. We also spent a lot of time improving the Streamlit interface so that it would remain simple and usable.
Accomplishments that we're proud of
We are proud that we were able to turn the initial idea into a working end-to-end prototype. Instead of building only an AI chatbot, we integrated voice input, AI explanations, voice output, quizzes, and visual learning into one application. We are particularly happy with the Hindi/Hinglish interaction and the combination of explanations with visual aids, because these were important parts of our original problem statement.
What we learned
The biggest thing we learned was that building an AI application involves much more than just calling an LLM API. We learned about prompt engineering, structured outputs, API limitations, speech-to-text and text-to-speech pipelines, error handling, fallback mechanisms, and designing AI features around real users. We also learned that making an AI system reliable often requires several iterations rather than expecting the first implementation to work perfectly.
What's next for VidyaSaarthi
We want to improve VidyaSaarthi beyond the current prototype by adding a proper RAG system using trusted educational/NCERT content, better personalization based on student progress, stronger regional-language support, and teacher-side features for monitoring learning. We also want to improve the voice interaction and make the system work better in low-connectivity classroom environments. Our long-term goal is to make VidyaSaarthi a practical AI teaching companion rather than just another educational chatbot.
Log in or sign up for Devpost to join the conversation.