· 9 min read
Thirty per cent of our volume stopped earning money
Yield was falling and every instinct in the room said send more. We sent less of what paid, and yield rose about eighty per cent.
Read itBuild log
What broke, and what I changed because of it.
Not advice. Incidents with dates on them, the theory I held before I understood the problem, and the rule I kept afterwards.
· 9 min read
The ad you cannot change after you send it
An image URL inside a delivered email is immutable, and Google caches it. Both facts are fixed. The ad still had to change on every open.
Read it· 8 min read
When the model is right and the number is wrong
Three places machine learning earned its place in an ad exchange. None of them were the clever part.
Read it· 5 min read
The lever was already in the data
Revenue went from ₹1.8 crore to ₹20 crore. The largest single contributor cost nothing to collect.
Read it· 7 min read
What one person and a model can actually build in eighteen months
Three production systems. Also: a list of the things that got no faster at all.
Read it· 6 min read
My RAG pipeline could not find an error code
Cosine similarity is very good at topic and very bad at identifiers. The fix was to stop choosing.
Read it· 6 min read
Six defects behind a green tick
An AI agent reported four phases complete. The widget had never once rendered.
Read it· 5 min read
Your firewall is lying to you
ufw said three ports were open. Seven were. Both statements were true at the same time.
Read it· 3 min read
Exit code zero is not evidence
A deploy that reported success every time and changed nothing. The signal everyone trusts is the one signal that cannot detect this.
Read it· 6 min read
The demo isn't the hard part
Prototypes got roughly ten times cheaper. Production did not get cheaper at all.
Read itThe build log, by mail
New notes, when I publish them.
One mail per note: what broke, the theory I held before I understood it, and the rule I kept afterwards. Irregular on purpose — I write these when something happens, not to a schedule, and I would rather send you nothing than send you filler.
Recognise any of these in your own stack?
That is usually the point at which a teardown pays for itself. Five days, fixed price, findings in writing.
Get in touch