Blog
Notes from the team
Thinking out loud about interpretability, control, and what we're building.
March 12, 20266 min read
Why every model needs a control layer
As models move into critical systems, observability isn't enough — teams need to steer and constrain behavior directly.
February 24, 20268 min read
Inspecting what your model actually learned
A practical walkthrough of using feature dictionaries to debug an unexpected production output.
January 30, 20265 min read
Shipping interpretability to production
The engineering decisions that let us trace live traffic without blowing up latency budgets.