llm-ops
Articles and projects by Daniel Chiwuzoh tagged llm-ops.
Articles
- How Confident Is the AI? Logprobs Explained ·
A forecast that says '70% chance of rain' is honest about its own uncertainty. Models carry that same number for every word. Here is how I read it.
- What Guardrails Really Mean (and How I Built Them) ·
Guardrails on a highway keep a car from going where it shouldn't. The same idea, applied to what a language model says, and how I built them into Ofuma.
- Don't Trust the Dashboard: How I QA My Own Software ·
A screen can render perfectly and still be lying to you. Three real bugs from Ofuma, and the habit that caught them: add the numbers up yourself.
- Prompts in Ofuma: Every Knob, and What It Actually Does ·
A prompt looks like a harmless little string until you try to change it in production. Every field, every parameter, and how to call it from four lines.
- Folders in Ofuma: Giving Ten Prompts a Filing System ·
Ten prompts in a flat list is fine. Sixty is a junk drawer. Nesting, paths, per-folder permissions, and moving prompts around in bulk.
- The Ofuma Dashboard: What Those Three Numbers Actually Mean ·
Almost no dashboard tells you how its numbers were worked out, which is when it starts lying politely. What Total Prompts, Active Prompts and This Month really count.
- The Features of Ofuma, and Why Each One Exists ·
You cannot ship code without version control, tests, and a way to roll back. Prompts are relearning that lesson. A tour of Ofuma, one problem at a time.