Building accounting software in public.

Notes from building Guliel, accounting and finance-operations software on one engine: what we shipped, what broke, and how the data model is put together.

• June 20, 2026·12 min read

  Production LLM Cost Optimization: Model Choice, Prompt Caching & Output Engineering

  Every LLM call in a hot path is a recurring bill, and cutting it usually costs you accuracy, speed or insight without anyone deciding to give them up. This is a practical guide to the four-way trade behind LLM cost optimization and the four techniques that move it on purpose (model choice, prompt caching, output engineering, batching), with a worked, measured example.

  llmcost-optimizationproductionai-engineeringprompt-cachingobservabilityinference
• May 18, 2026·90 min read

  Software Architecture Has No Rigor, and AI Feeds on It

  Every serious engineering field can verify a design before it is built. Software architecture cannot, and an AI generating against an unverifiable spec will hallucinate, duplicate, and reorder with impunity. Here is the method I use at Guliel: model software as what it is made of (data, transformations, and a place to run) in a diagram with rules you can check.

  architecturesoftware-designcategory-theoryaiengineering