BookStream: Voice AI for Literary Discovery

December 3rd, 2023

Bookstream, the TLDR

BookStream is a voice-driven application that provides personalized book recommendations. This solution offers a unique, hands-free way to discover new books tailored to users' preferences.

Users call in, mention their favorite book, and get similar book suggestions through a seamless integration of Weaviate's vector search, Pulze.ai's model selection (optimizing for speed), Twilio's Voice API for phone calls, and Eleven Labs' TTS API, all hosted on Replit.

Problem Addressed

BookStream leverages Vector Search to address the challenge of finding relevant and engaging new books in a vast and ever-growing literary landscape. It solves the problem of choice overload and the time-consuming process of searching for books that align with a reader's personal taste.

Accessing the Project

Public Replit Link (open-source):

Installation on Replit

Fork the Replit: Visit the Replit project URL and click on 'Fork' to create your own copy of the project.
Install Dependencies: Dependencies are automatically handled by Replit; ensure the replit.nix file includes all necessary packages.
Set Up Environment Variables: Use Replit's Secrets tab to add environment variables for Twilio API keys, Weaviate endpoint, Pulze.ai credentials, and Eleven Labs API Key.

The Journey of a Agentic Voice Request

Initiating the Interaction: The process begins when a user articulates their favorite book into the system. This can be done anywhere, anytime, just by using their voice over the phone.
Vector Database Search: Upon receiving the user's input, the application uses a vector database, powered by Weaviate, to find similar books. This database runs a semantic search and intelligently analyzes the content and context of the user's favorite book to identify other books with similar themes, styles, or authorship.
Language Model Processing: The suggestions from the vector database are then ingested by a sophisticated language model. Pulze.ai is used to choose an appropriate language model to query, optimized for speed, which is critical given the voice interface for this application. It not only lists the suggestions but also provides a brief rationale for each recommendation, enriching the user’s experience.
Text-to-Speech Transformation: Once the response is crafted, it is transformed from text to speech using the 11 Labs API. This step is crucial as it maintains the natural, conversational tone of the response, making it more engaging and personal for the user.
Delivering the Recommendations: Finally, the spoken recommendations are delivered directly to the user via Twilio's robust communication API. This ensures a seamless and real-time interaction, almost as if the user is having a conversation with a knowledgeable friend.

The Technology Stack
- Twilio account for API integration.
- Weaviate vector search engine access.
- Pulze.ai account for speed optimization.
- Eleven Labs access for voice.
- Replit account for hosting and running the application.
Our Principles for a Great Voice UX:

Simple and Accessible: Just speak and receive recommendations.

Personalized Responses: Tailored suggestions with explanations for each book.

Conversational Tone: Maintains a friendly and engaging interaction.

Conclusion

We view BookStream is a new chapter in the story of book discovery. By harnessing the power of voice AI, vector databases, and advanced text-to-speech technologies, it offers a uniquely intuitive and engaging way for readers to find their next favorite book. Whether you're a casual reader or a literary enthusiast, BookStream is here to guide you through the vast world of literature, one voice command at a time.