Sonara

• 

• 

• 

• 

  End-to-end architecture showing how Google Gemini powers each Sonara feature.

Inspiration

I come from Bihar, a state in India where many villages and small towns still lack access to good English coaching. When I was younger, I studied in such an environment and didn’t have many opportunities to practice speaking English properly. Later, my family moved to another state and I joined a private school. The students there were much more confident in speaking English. Their pronunciation and fluency were much stronger, and I often felt left behind. At times, my pronunciation was even made fun of, which made speaking even more intimidating.

Most people know english but they feel under confident due to their accent, pronunciation, and they lack a safe space to practice speaking and improve pronunciation. That is how Sonara was born: an idea to create a friendly AI system that helps people practice speaking without fear or judgment.

What it does

Sonara is an AI pronunciation coach that helps users improve their spoken English through real-time feedback. Users can:

• Talk 1-on-1 with an AI coach that listens to their speech and gives pronunciation guidance.
• Read along with text (chapters, articles, or custom paragraphs) while Sonara analyzes their pronunciation.
• Practice with friends in a shared session, where users talk together and the AI highlights mispronounced words. The goal is to make pronunciation learning interactive, social, and confidence-building rather than intimidating.

How we built it

Sonara is built as a web-based application combining real-time audio processing with AI language analysis.

• The frontend captures speech through the browser microphone.
• Audio is streamed to the backend using WebSockets for real-time communication.
• The speech data is processed using Gemini API, which generates transcription and feedback.
• The AI analyzes the spoken words and highlights possible pronunciation mistakes.
• The frontend interface displays this feedback through a conversational chat-style UI.

Challenges we ran into

One of the biggest challenges was implementing real-time communication using WebSockets, especially for the feature where two users can speak with each other in a shared session. Handling voice input, sending audio data to the backend, transcribing it, and generating AI feedback simultaneously required careful synchronization. At times, the transcription and AI responses were delayed or out of order. Ensuring that both users could interact smoothly while the AI processed speech in real time took significant debugging and experimentation.

Accomplishments that we're proud of

We are proud that we were able to build a working AI-powered pronunciation coaching prototype. Some key accomplishments include:

• Creating a real-time AI speaking coach that listens and provides feedback
• Implementing speech transcription and pronunciation highlighting
• Building a read-along practice feature for improving fluency
• Designing a shared practice room where users can speak and learn together

What we learned

While building Sonara, we learned a lot about real-time AI applications and speech processing systems. We gained experience working with:

• real-time audio streaming
• WebSocket-based communication
• integrating AI APIs for speech analysis
• designing user interfaces that clearly present AI feedback
• deploying backend services using Google Cloud Run

What's next for Sonara

Due to joining the hackathon a bit late, I could not fully achieve the desired functionality for the shared speaking rooms. In the future, I plan to expand Sonara by introducing a swipe-based matching system, similar to dating apps, where two anonymous users can connect and practice speaking with each other. Future improvements include:

• enabling global speaking rooms where learners from different countries can practice together
• supporting multiple languages and regional accents
• adding accent detection and personalized pronunciation coaching

Sonara is not just a tool for learning pronunciation it’s a space created where people can practice speaking without fear

Built With

• gemini-2.5-flash
• node.js
• react
• typescript
• websocket