Case Study: Tavus achieves human-grade conversational AI with Qdrant Edge

Tavus cuts retrieval latency to 20–25ms with Qdrant Edge

Tavus, a human-computer research lab building a Conversational Video Interface (CVI), faced a challenge in grounding their AI conversations with private customer knowledge. Adding a retrieval (RAG) layer introduced network latency that threatened to push their utterance-to-utterance response times beyond the subsecond threshold required for natural human conversation. They needed a fast, local retrieval solution to maintain their performance standards.

The solution implemented with vendor Qdrant involved using Qdrant Edge to create per-conversation vector stores colocated with their conversational workers. This eliminated network hops, reducing retrieval latency to 20-25ms. As a result, Tavus achieved best-case end-to-end response times of 500-600ms, indexed millions of data points seamlessly, and allowed customers to onboard private knowledge without building their own RAG systems.

Tavus