<ul><li>Mastercard achieved a 400% improvement in latency, accuracy, and cost-effectiveness using Opensource LLMs, Elastic search, and FIASS.</li><li>The project involves the automatic matching of merchant names in financial records to standardized business names to facilitate spendings sorting, financial analysis, fraud detection, and clearer transaction histories for customers.</li><li>The project is structured into three stages: parsing messy transaction text, searching for the most likely business match using a hybrid of ElasticSearch and FIASS, and selecting the best match using large LLMs and an 'AI judge.'</li><li>Breaking down the problem into clear stages like parsing, hybrid search, and re-ranking has shown that building powerful AI systems is achievable even with modest resources, emphasizing thoughtful design and smart tool utilization over raw compute power.</li></ul>

Gen AI Merchant Matching — using Opensource LLMs, Elastic search and FIASS

Discover more