SPEEKAR

Intelligent and Cost-Effective Knowledge Retrieval

Speekar is a cutting-edge platform designed to enable quick and low-cost knowledge retrieval without hallucinations, utilizing a combination of structured and unstructured data.

It employs an innovative keyword-based search framework to reduce the time and cost of using large language models (LLMs). The system uses a smaller LLM to generate and cache keywords for efficient context discovery. A larger LLM is then employed to provide tailored answers using the identified context. This framework significantly reduces inference time and cost, making it ideal for commercial chatbot and search applications.

Speekar also integrates a speech-based interface for seamless interaction, allowing users to input queries and receive spoken responses, combining the power of advanced AI with user-friendly accessibility.

Research Paper

Learn more about the research and framework behind Speekar:

Codebase

Access the Speekar project's codebase on GitHub:

Team Members