I’ve spent the last few years moving between two worlds: research into how meaning gets inferred in language, and the day-to-day reality of shipping ML systems that real teams depend on. This blog is where those two threads meet.
What to expect
I’ll be writing about explainability methods that hold up in production — not just SHAP values on a slide, but the engineering work of monitoring explanation drift, building audit trails, and making models legible to the people who have to answer for their decisions.
I’ll also write about the research side: computational pragmatics, natural language understanding, and the parts of meaning that models still struggle to infer.
Why now
I’m currently building Darsha, an SDK for ML explainability and compliance. Writing here is partly how I’m working through the ideas behind it, and partly an open notebook for anyone tackling the same problems.
More soon.