When RAG Can`t Be Wrong

We will have an unmissable chat in Portuguese with Alex, AI Engineer straight from Frankfurt! He will present a practical experiment and answer a fundamental question for those building AI systems: Is it possible to abandon expensive commercial AIs for complex tasks and migrate the pipeline to local hardware?📌 What you will see in the talk:Real Case Study: How we took Marcus (RAG prototype with agents for legal research in Latin America, originally made with Gemini and Claude) and tested its limits by cloning it on Gemma 4 via Agent Development Kit (ADK).Infrastructure: Running everything on a personal-use GPU (RTX 4070 Ti Super 16GB).Engineering in Practice: Structuring prompts, optimizing token budget, and local evaluation to bring open models closer to major commercial APIs.Results and Challenges: Obstacles in logical formatting, real learnings, and a clear roadmap for those looking to migrate agent prototypes to open-weights environments.