Notes from production systems — multi-tenant SaaS, AI pipelines, reconciliation engines, and the database designs underneath them.
Most billing systems store a balance column and slowly corrupt it. Here is the double-entry model I used for a municipal water utility, and why arrears and partial payments force your hand.
Sydney and Zurich are nine hours apart. Serving both doesn't mean someone takes the bad call — it means synchronous time is scheduled where it's civil for everyone, and the rest of the work doesn't need it.
How I built an offline-first mobile API for municipal water meter readers who work in areas with no connectivity — the sync model, the conflict rules, and what a queued write actually has to carry.
Inference providers go down, rate-limit, and degrade. I run a failover chain across SambaNova, Groq, Gemini, OpenAI, and Cerebras — here is why a retry loop is not enough and what a circuit breaker actually buys you.
Exact match fails on real documents. Here are the eight comparison algorithms and three matching strategies I use to reconcile invoices, delivery notes, and payment terminal settlements across four companies and eight stations.