A few of the rooms we have worked in.
Each one written up as a case study — the problem as we found it, the constraint we worked under, and the change that held. Client names are withheld; the numbers are not.
Cutting p95 checkout latency to 340 ms
A checkout that took 1.4 seconds at p95 was losing orders at the last step. Two slow spans accounted for 93 percent of it. Neither was the database.
Read engagement →
Cutting unsupported answers from 9% to 0.4%
A support assistant was answering confidently and wrong, inventing policy details that were never in the docs. The model was fine. The chunker was splitting policy tables mid-row, so the row the model needed never reached it.
Read engagement →
From 41 minutes to 6, and nobody is afraid anymore
A deploy took 41 minutes and everyone dreaded it, so the team batched a week of changes and shipped on Fridays. Three slow steps were doing the damage. None of them needed a new tool.
Read engagement →