Most of us have interacted with Large Language Models (LLMs) like ChatGPT or Gemini through standard browser chat windows. However, this simple approach often leads to concerns about data privacy, security, and ongoing subscription costs. Furthermore, many of us have experienced model "hallucinations" or generic outputs that do not align with our specific appraisal processes, making AI results difficult to integrate safely into jurisdictional workflows.
This presentation addresses both of these critical issues. First, we will introduce a secure, open-source alternative: running powerful models locally on standard hardware using Hugging Face. This approach keeps your sensitive data entirely on-site and free from recurring API fees. Second, we will demystify how to use LangGraph to build structured, stateful workflows. Think of this as creating a custom map for your AI to follow. You will learn how to tightly control every step the model takes, dictate its actions after a decision is made, and choose exactly what tools (like local databases or calculators) it can access to aid its reasoning.
This session features an accessible introduction to these two open-source frameworks, followed by a step-by-step demonstration of a LangGraph agent facilitating daily efficiency in mass appraisal. Attendees will walk away with a practical understanding of local LLM's to help their offices start charting their own path and uncovering the hidden "rich stuff" in their data.
Joshua Jorgensen, Machine Learning Engineer, Thimgan & Associates, FL
Keywords: Artificial Intelligence & Machine Learning, Mass Appraisal & Valuation Modeling, Advanced
