Utilizing Open Source Models to Build AI Agents for Industry

This talk explores how to use open-source LLMs models to build AI Agents to enhance productivity and efficiency across various industries. As AI Agents become a major focus by 2025, a wide range of applications will be widely promoted, providing enterprises with opportunities for challenge and upgrade. Open-source models have three main advantages: on-premises deployment enhances security protection, ownership of weights increases system stability, and fine-tuning creates a better user experience. Moreover, the open-source Breeze 2 from the MediaTek Research not only serves as a Traditional Chinese model but also targets AI Agents; we will open source a lightweight Agentic Workflow to help Taiwanese industries easily deploy Breeze 2 as an industrial assistant. This presentation will illustrate how to use these open-source models to create AI Agents through examples, including real-world cases from Cola Tour and AIA.

這個講題探討如何利用開源的 LLMs 模型來建立 AI Agent，以在各行業中提升生產力和效率。隨著 2025 年 AI Agent 成為一大重點，各類應用將廣泛推展，企業因此迎來挑戰與升級的契機。開源模型具備三大優勢：地端部署提升資安防護、擁有權重增加系統穩定性、以及可微調創造更佳使用者體驗。此外，聯發創新基地開源的 Breeze 2 不僅是一個繁體中文模型，更是一個對標 AI Agent 的模型；我們將開源一個輕量級的 Agentic Workflow，以助力台灣產業輕鬆部署 Breeze 2 成為產業助手。本演講將通過實例，包括使用可樂旅遊和 AIA 的實際案例，說明如何利用這些開源模型打造 AI Agent。