Generative AI Merchant Matching - Building Cost-Effective Enterprise AI Systems

Learn how to build cost-effective enterprise AI systems for merchant matching in this 20-minute conference talk from Databricks. Discover a three-step approach that uses fine-tuned LLMs and advanced search techniques to match merchant descriptors to known businesses, achieving results comparable to expensive alternatives at minimal cost. Explore how a fine-tuned Llama 3 8B model parses merchant descriptors into standardized components, while a hybrid search system uses these components to find candidate matches in a database. Understand how a Llama 3 70B model evaluates top candidates with an AI judge reviewing results for hallucination detection. Gain insights into achieving 400% latency improvements while maintaining accuracy and keeping costs low, with each fine-tuning round costing only hundreds of dollars. Master key concepts in prompt engineering, fine-tuning strategies, and cost and latency management through practical examples that demonstrate how small teams with modest budgets can effectively tackle complex AI problems using careful optimization and simple architecture for balancing cost, speed, and accuracy.