Yoni Michael

Before Typedef, Michael worked on infrastructure at Salesforce. A Salesforce engineering account of Dreamforce 2017 identifies him with work on predicting hard-drive failures. His team had to clean and standardize metrics from different drive vendors before using machine learning to associate conditions with failures. That earlier work illustrates a recurring concern in his current writing: a capable algorithm still depends on a usable representation of the system it operates on. Typedef lists Michael and Kostas Pardalis as co-founders.[3][1]

Typedef also publishes fenic, a semantic DataFrame engine. It combines familiar filtering and joining with model-backed extraction, classification and summarization in inspectable pipelines. Those pipelines can be rerun and examined instead of leaving the result as an unstructured chat session. This is company work; Michael’s verified connection is as co-founder, not sole author of the codebase.[4][1]

In their data-context essay, Michael and Pardalis use a join-key change to explain the gap between reading SQL and understanding a data system. A syntactically valid change can alter the grain, the unit represented by each row, causing an aggregation to double-count revenue further downstream. Their context layer is intended to connect code, lineage, metric definitions and dependencies so an agent can examine those consequences. The contribution is a system model around the agent, rather than a claim that better text generation alone solves data correctness.[2]