A VIABILIDADE DE UMA ARQUITETURA HÍBRIDA DE CHATBOT COM RASA E LLMS LOCAIS
A VIABILIDADE DE UMA ARQUITETURA HÍBRIDA DE CHATBOT COM RASA E LLMS LOCAIS
-
DOI: https://doi.org/10.22533/at.ed.1281125170311
-
Palavras-chave: Rasa, Large Language Model (LLM), Chatbot, Arquitetura Híbrida.
-
Keywords: Rasa, Large Language Model (LLM), Chatbot, Hybrid Architecture.
-
Abstract: In a scenario where the use of artificial intelligence is growing exponentially, especially in corporate and industrial environments, ensuring control, security, and cost efficiency is essential. This paper proposes and evaluates an innovative hybrid architecture for chatbots, utilizing the Rasa framework and applying its smaller, personalized, and controllable language model to act as a manager and filter for a more complex and robust Large Language Model (LLM). The proposal is to use Rasa as the main user interface, responsible for interpreting commands and applying rules, delegating tasks to the LLM only when the complexity of the request requires greater language understanding. The technical feasibility of the approach was validated through accuracy and latency tests, which, although revealing considerable latency in a development environment, confirm the architecture's potential for a fixed-cost model and its applicability in various practical scenarios.
- Sergio Juniors Garcez
- Joethe Moraes de Carvalho