One of the dirty little secrets about large language models (LLMs) is that they can’t easily interpret structured information captured in data warehouses and databases (or even CSV files, really).
Yes, all that work to create schemas and structured datasets to reflect the business processes — all for naught, when it comes to powering an LLM at any rate.
Now Kumo, a San Francisco startup, is moving beyond the LLMs to offer greater discoverability in structured and semi-structured data, the mainstay of the enterprise.
The company has…