Engineering notes from Avectic. We build deterministic decision infrastructure for AI-assisted workflows. These posts document the architectural patterns, design choices, and field lessons from that work. They are written for engineers, product leaders, and anyone thinking seriously about how AI should enter consequential decisions.
Ask a language model the same question twice and you may get two different answers. That property is fine for a chat interface and disqualifying for a regulated decision. This post describes the architectural pattern that separates probabilistic interpretation from deterministic decision, and the design choices that make that separation work in production.
Read →If you are building a deterministic decision engine, the most important design decision you will make is how you represent rules. This post describes the Criterion object pattern we use to encode 120+ healthcare payer policies and 67 NCCI bundling rules, the composition layer on top of it, and the three mistakes most teams make on their first rule packs.
Read →The AI industry is shipping consequential-decision systems built on probabilistic components and papering over the reliability gap with disclaimers, explainability tooling, and human-in-the-loop review. None of these substitutes for the deterministic evaluation layer that regulated decisions actually require. This post argues the pattern is not optional, and the regulatory environment is catching up fast.
Read →Most audit trails are designed for the institution that owns the system. A small architectural shift—building them for the person on the receiving end of the decision as well—changes how decision systems behave in dispute, in appeal, and in remediation. This post describes the shift, why it is not more expensive to build, and why the default design choice is worth questioning.
Read →Eighteen months of building a deterministic decision engine for spine surgery prior authorization produced lessons that do not appear in most AI industry discussions. The rules are harder than the AI. The domain experts know things AI teams underestimate. Determinism is a trust property, not just a technical one. This post documents what we got right, what we got wrong, and what generalizes.
Read →