Conversational AI Voicebot

AI Agent System that user can call a regular telephony number and have conversation with an AI agent for their queries regarding the order status of the product they purchased or the product details for easy product buying recommendation according to the user’s needs. Voicebot will ask for specific user needs based on the product catalog provided to AI Agent and can then suggest product recommendations. Enhancing the user buying experience for particular ecommerce store.

Scope

The AI agent is designed to perform rapid product searches and provide detailed recommendations based on the specific context of what the customer is looking for. By integrating Generative AI, the store seeks to reduce the effort and time required from customers during support interactions. Additionally, the AI system will enable the simultaneous handling of multiple customer queries through a single telephone number, further streamlining the support process and improving overall customer satisfaction and operational efficiency.

Our Approach

We leverage cutting-edge WebRTC technology provided by Livekit to manage a high volume of concurrent calls efficiently by employing agent-worker isolation. This ensures that each participant (caller) is handled by a dedicated worker. We use Livekit’s room service to segregate participants and workers into separate rooms, simplifying backend infrastructure management. The agent connects to telephony via SIP, facilitated by Twilio and Livekit. When a user (participant) calls a Twilio number, the request is routed to the SIP server, a room is created, and both the worker and the user join this room seamlessly.

To deliver a fast and responsive experience in the Generative AI era, we use advanced speech-to-text (STT) technology from Deepgram. This ensures minimal latency and efficient interaction. We also offer flexibility for customers to choose their preferred language models and text-to-speech (TTS) modules based on their needs. 

We have also integrated a Retrieval-Augmented Generation (RAG) system to effectively address specific user queries. Furthermore, a live function call feature has been incorporated to retrieve order status, which is seamlessly connected to the eCommerce store’s API.

Tech Stack