VoiceRouter AI

Inspiration

In today's digital-first world, navigating through customer service departments can be a frustrating experience. Traditional IVR systems are rigid and menu-driven, while chatbots often miss the human element. We wanted to create a more natural, accessible way for users to get connected to the right department simply by speaking their needs.

What it does

VoiceRouter AI is a web application that uses voice recognition and natural language processing to understand user inquiries and automatically route them to the appropriate department. Users simply click the "Ask AI Assistant" button, speak their question or concern, and our AI instantly analyzes their request and directs them to the most relevant department, complete with confidence scores and matching criteria.

How we built it

Our application is built with a modern tech stack focused on performance and accessibility:

Frontend: React with Vite for a lightning-fast development experience TailwindCSS for responsive, beautiful UI components Web Speech API for native speech-to-text capabilities Dark mode support for accessibility

Backend: Python-based classification service Natural Language Processing for query understanding RESTful API architecture for seamless integration The application flow:

User speaks their query through the browser's native speech recognition Real-time transcription appears on screen The transcript is sent to our classification service AI analyzes the query and determines the most appropriate department Results are instantly displayed with confidence metrics

Challenges we ran into

Cross-browser Speech Recognition: Initially, we started with the Vosk speech recognition library for broader browser support. However, we discovered that the Web Speech API provided a better user experience for modern browsers, leading us to pivot our approach.
Real-time Feedback: Balancing between showing interim results and waiting for final transcriptions required careful consideration of user experience.
Classification Accuracy: Fine-tuning the backend classifier to handle the wide variety of ways users might phrase their questions was a significant challenge.