A top-3 national retail bank ran daily teller operations on slow, archaic branch software with no UX. The Angular replacement was a one-feature prototype when I joined. I grew into owning technical direction: hired 2 through a loop I wrote when an engineer left, led 3. State drifted and caused bugs, so I standardised on RxJS, already in Angular and enough, not NgRx, and on Sentry before rollout.

Nothing cut from the planned scope, release-blocking issues caught before rollout, and the app took over daily teller operations across 500+ branches.

People write freely, but the text has to publish without revealing who wrote it — an employer mentioned in passing, a personal domain, a contact detail. Pattern matching caught the obvious forms and missed anything phrased casually, so I put a small-model classifier in front of every free-text field, gated by a golden eval set of hand-verified cases including prompt-injection attempts.

Leaks are caught before anything is visible, and a prompt or model change that regresses fails the eval instead of reaching a person.

An enterprise fintech ran 20+ payment apps locked into shipping together by co-deployed frontends. The coupling sat in the bundle, not the pipeline, so automation could not fix it. In a 3-engineer specialist team I co-designed independently deployable micro-frontends behind a shared shell, built a working MVP to prove it, and set the 6-month roadmap to break that coupling.

Approved and funded at SVP level, then adopted and carried forward by another team.

A 35,000-line Angular module in the legacy console sat on the first-load path hit by every customer. A rewrite meant months of work, an undefined regression surface, and two stacks through cutover, and a broken path here is a paying-customer incident. I won leadership backing for incremental code-splitting instead, sequencing it so each slice landed a win and could be reverted alone.

Page load down 48% and DOM Content Loaded down 53%, with zero production regressions across the rollout.