Integración de modelos de lenguaje avanzado para la mejora de gestión de conocimiento del chatbot de PlanEsTIC

Abstract

The chatbot of the PlanEsTIC area at the Universidad Distrital Francisco José de Caldas (UDFJC) faces critical limitations in its current operation. By relying on a static knowledge base integrated with Cohere and AWS Bedrock, the system generates outdated responses, lacks access to real-time information, and shows an inability to handle complex or open-ended questions, leading to a frustrating user experience. To overcome these issues, this internship project focuses on redesigning the chatbot's architecture through the integration of the DeepSeek language model. The overall objective is to expand its capabilities toward a multimodal system that, leveraging DeepSeek’s contextual reasoning and web access, can accurately respond to any type of query related to PlanEsTIC and the UDFJC, while maintaining coherence with the institutional information available to the university community.