Business Strategy Copilot: uma prova de conceito de assistente inteligente usando modelo small language model (SLM) para o planejamento estratégico de startups

Abstract

Large Language Models (LLMs) are modern natural language processing models that are behind technologies such as Gemini (Google), Copilot (Microsoft), and ChatGPT (OpenAI). However, LLMs can often have high computational and energy requirements, making the development of solutions based on these technologies dependent on large technology companies. In this context, the main objective of this work is to explore Small Language Models (SLMs), specifically the pre-trained Phi3-Mini model, as a relevant and efficient alternative, capable of providing intelligent responses to complex problems and with lower resource demands. Thus, in a Python programming environment supported by the open-source Ollama platform, configured on a Dell Inspiron 155510 laptop, an application was developed to assist the strategic planning of startups. The Python application combines good practices in object-oriented programming and prompt engineering, allowing the generation, refinement, and prioritization of strategic objectives based on data such as the startup's mission, vision, and customers, applying quality control to the responses. A case study demonstrated the efficiency of the model on local hardware, and all source code was made publicly available on Github, encouraging the use of AI with lower computational and energy costs for management, entrepreneurship and innovation purposes.