Writing
Field notes
The decisions behind real systems — what we tried, what failed, and the principle worth carrying into the next build. We write only what we can stand behind, including where an approach did not work and where our own role ended. For the short version of any of these, see the case studies.
-
Field note · Production AI
Two paths to a trustworthy answer
Match rigor to the cost of being wrong.
How we gave business users plain-English access to a hundred-table warehouse — by splitting the system by cost-of-wrong: an exploratory path where the user verifies the answer, and a deterministic path built on vetted queries.
-
Field note · Production AI
Optimize the reviewer, not the robot
When a human checks everything, optimize the check.
A lawyer reviews everything regardless — so the win was never machine autonomy. How a legal document-review system reached production by making the human's review effortless, and choosing the simpler architecture every time.
-
Field note · Cloud & security architecture
The layer underneath the data
The platform is everything around the data, not the data.
Five-plus siloed warehouses became one governed platform. We owned what sat underneath and around the data — hosting, security, governance, cross-cloud wiring — and how the architectural know-how stayed in-house when the build was done.
-
Field note · Production AI
Build the engine, not the feature
Build the reusable capability, not the point tool.
An R&D notebook prototype became a production knowledge engine. The defining decision was refusing to build a point tool — the same reusable core went on to draft tender responses and generate slide decks.