No tutorials on things I have not run. Every article here comes out of a decision I had to make, defend, and then live with in production.
A till that stops working when the internet does is a till that loses money. Here is the outbox-plus-idempotency contract that lets a Tauri desktop POS sell offline and replay safely into a Django backend - without double-charging anyone.
ReadSeams before services. How to extract modular services from a legacy monolith incrementally - with the data ownership rule, the strangler pattern, and an honest list of what you give up the moment a function call becomes a network call.
ReadHow FOR UPDATE SKIP LOCKED plus expiring leases gives you a distributed, crash-safe work queue without adding a broker to your architecture - with the reaper, backoff and priority scheduling that make it production-grade.
ReadA cost-ordered cascade - URL hash, content hash, SimHash banding, then BGE-M3 semantic comparison - that kills the vast majority of duplicates with an index lookup and reserves embeddings for genuinely ambiguous pairs.
ReadRedis Streams consumer groups give you delivery guarantees; they do not give you correctness. What idempotent handlers, delayed retries, dead-letter streams and stale-message reclaim actually look like in production.
ReadRow-level tenancy is one missing WHERE clause away from a data breach. Schema-per-tenant moves isolation from something developers remember into something the database enforces - and here is what it costs.
ReadConnection exhaustion, N+1 queries, missing composite indexes and recomputed dashboards - the four things that actually limited a Django SaaS, and why PgBouncer came before hardware.
ReadWhat changes when the same backend has to serve a React dashboard, a React Native app and a biometric device on a gym LAN - untrusted clocks, offline buffers, and why the hardware client reshaped the API contract.
ReadThe single boundary that separates an LLM feature you can point at production data from a demo you cannot - plus schema-validated extraction, provider fallback and cost accounting as engineering concerns rather than prompt concerns.
ReadEvery manual step in a release is a step someone will skip at 11pm. How tag-driven CI, generated changelogs, semantic versioning and Poetry lockfiles turned publishing a Python package into a single command.
Read