Embeddings in the Real World: What Works & What Breaks

Hack Session

About the session

Embeddings are amazing at transforming messy text into mathematical meaning. But anyone who has tried to match a customer problem to the right solution, find a supplier in a database, or map a keyword to a commodity knows that embeddings start to struggle when confronted with real‑world data. Short words, ambiguous terminology, numeric fields, taxonomy constraints, and inconsistent formats—this is where semantic theory meets the chaos of enterprise reality, and a similarity calculation alone is rarely enough. 

In this talk, we show what embeddings actually do, where they shine, and—more importantly—where they break down. This isn’t just about cosine similarity or vector search; it’s about the engineering decisions required to bridge the gap between user queries and messy, structured business data. 

Session Takeaways

    • A clear understanding of where embeddings add value in text‑similarity and matching problems. 

    • Insight into why hybrid systems outperform “embeddings‑only” approaches. 

    • An honest look at the practical limitations of embeddings when handling hierarchy, numeric logic, and business constraints. 

    • A sense of when embeddings are the right tool—and when they are not.

    Speaker

    Download Brochure

    We use cookies essential for this site to function well. Please click to help us improve its usefulness with additional cookies. Learn about our use of cookies in our Privacy Policy & Cookies Policy.

    Show details