Problems worth solving properly.

01 / Enterprise architecture

Salesforce design authority

Problem. Multiple delivery teams needed a consistent architectural standard for changes to a shared enterprise Salesforce platform.

My role. Lead architect responsible for platform design authority, reviewing solution approaches and setting standards and patterns for a team of roughly 30 people.

What I did. Made architecture decisions explicit, challenged designs before implementation and gave teams reusable patterns rather than solving the same class of problem repeatedly.

What it taught me. Quality improves when important decisions become durable constraints instead of relying on somebody remembering the right answer during every review.

02 / AI-assisted engineering

Making model output easier to trust

Problem. LLMs can produce strong engineering work and then fail on something that looks trivial, while sounding equally confident in both cases.

My role. Architect and builder exploring how to use model reasoning without making the model the final authority for repeatable engineering rules.

What I did. Separated open-ended reasoning from deterministic checks, durable state and independent evidence.

What it taught me. The useful question is often not how to make the model smarter. It is which part of the job should stop being probabilistic once the rule is known.

Read: Determinism over probability

03 / Delivery assurance

Turning “looks right” into evidence

Problem. Reviews become weak when completion depends on a person or model saying the work appears correct.

My role. Solution architect defining delivery controls and the evidence required before work can be treated as complete.

What I did. Moved stable checks into repeatable validation and separated the person doing the work from the person or mechanism accepting it.

What it taught me. Confidence is useful during design. Evidence is what should close the work.

See: Agent Learning Loop