11th International Conference on ICT for Sustainable Development

Authors - Ajuram. P, E. Grace Mary Kanaga
Abstract - Large Language Models have demonstrated great effectiveness in generating text and images. However they can become even more efficient by perfecting the prompt given to them. This paper proposes a multi LLM framework that dynamically orchestrizes several specialized LLM models in accordance with complex user prompts. First, a primary LLM analyzes the user prompt and breaks it down into multiple sub tasks. Then, for each identified sub task with respect to its type (text to text, text to image, or image to text), a suitable LLM is assigned. The context, instructions, and the output format is also provided by the primary LLM for each chosen model. The sub tasks are executed either in parallel or in sequential order. This approach automates the workflow, optimizes model utilization, and improves response relevance, making it suitable for applications requiring multi modal collaboration and processing.