NeuralQ Engineering
Own Every Layer
Architecture decisions and the reasoning behind them, engineering economics with the real numbers, and what it actually takes to run production without letting a vendor’s pricing page decide your roadmap.
No launch announcements. No thought leadership. Decisions, tradeoffs, and cost tables. The beliefs behind all of it are at /beliefs.
Sixty-two migrations and no rollback button
Schema change as a reviewed artefact: why we write SQL by hand, what an immutable migration really means, and the ordering bug that made every deploy exit 1 at boot for weeks.
Read →A request id that survives the boundary
Correlation across a Java core, a Python service, and a job queue — the two-key trick, why the id is captured at enqueue rather than claim, and the filter that silently ran twice.
Read →Everything outside the core must be replaceable — and what that rule cost us
Our one non-negotiable engineering rule, the test that tells a real seam from indirection, the two interface shapes we got wrong, and an honest accounting of what it slowed down.
Read →Running production at zero fixed infrastructure cost
Not a free trial and not a credits programme. The architecture that makes it survivable, the cold-start numbers we measured, the free-tier ledger we keep, and the tradeoffs we accepted rather than hid.
Read →No Kafka in V1: the complexity we refused to buy
A modular monolith and a Postgres-backed queue, with the actual claim query, the retry and reaping mechanics, the four failure modes this design really has, and the measured condition that would change our minds.
Read →Four languages, one domain: why the stack is polyglot on purpose
Java for the core, Python for AI orchestration, TypeScript at the surface, SQL as the substrate — the boundary discipline that stops that becoming chaos, and the bill we pay for it.
Read →