Proud to present a chatbot developed to assist in data trading- both in marketing data and sourcing third party data. The system was designed and implemented as part of Ignacio Moyano Fernández’s BS Thesis. Following a nice concise presentation, Ignacio obtained a well-deserved 9,5/10.
This intelligent conversational assistant is based on the Retrieval-Augmented Generation (RAG) architecture and is designed for integration in data marketplaces and data spaces. It allows users to query both a theoretical knowledge base about data markets and a catalog of commercial products using natural language. It works in several languages and its responses are traceable and grounded in real data.
Watch the English demo video here or pulsa aquí para la demo en español.
The chatbot is entirely built using open-source components and open LLM and embedding models and runs on regular consumer desktop hardware. This ensures complete control over the confidentiality of the underlying data.
Our findings indicate that injecting knowledge using RAG significantly reduces hallucinations and ensures the faithfulness and relevance of responses. Additionally, this approach allows smaller more resource-efficient models to almost match the performance of larger, more complex ones.
Many thanks to my previous student, Miguel Eleno, who collected the data this work relies on. This is another brick in the Data Economy Observatory at UPM. I am very much looking forward to welcoming new students to help build this system that will decisively contribute to increasing the transparency of emerging data markets.
Santiago Andrés Azcoitia 
Deja un comentario