LaVague Webinar: how to build and improve Large Action Models using LLMs

​​​🌊LaVague (https://github.com/lavague-ai/LaVague) is an open-source Large Action Model framework to build AI Web Agents. Our framework makes it easy to design agents to perform actions for us on the web, by piloting a driver using Large Action Models.

​We have two core components to make it happen:

• ​Our World Model: powered by OpenAI GPT-4o multimodal to reason using the user’s objective (Fill this form) on the screenshot of the web driver and provide instruction (Click on Apply button) to our action engine.

• ​Our Action Engine: powered by Llama Index to take the instruction of the World model to turn it in Selenium code. 

​In this webinar, we will dig deeper into the inner workings of our Action Engine to understand how we designed our RAG pipeline to consistently produce the right code to interact with the current page’s web elements.

​We will show insights and tips on how to improve the retrieval stage on the DOM, and how to best generate the right Selenium code using the retrieved HTML associated to the user instruction.

​We will do a demo, present our architecture and code, explore evaluation and share potential next steps to improve our framework as a whole, along with tips for users to customize their Large Action Model.

​🚀So if you are interested in building performant Agents, are interested in Large Action Models, or want to contribute to our open-source framework, don’t hesitate to drop in!