Amir Cabili

A 30–45 minute discovery call covers the product, constraints, current stack, and definition of done. I follow with a short plan: milestones, risks, and the first piece of work. If I am not the right fit, I say so.

Both. At Hub Security I lead Angular 20+ product work and ship React projects in the same role. At eToro Marketing I delivered React projects as well as Angular 16+ work. React is also a certified MERN practice (React, Node, Express, MongoDB).

I start from the hiring market, the existing codebase, and the type of interface. Long-lived enterprise systems with strong typing often fit Angular. Products that need a thinner runtime, a React-shaped team, or a Node-native stack often fit React.

At Hub Security it means OpenAI APIs in the product architecture: prompt systems, copilots, AI-assisted UI generation, automation in internal tools, and interface states for loading, refusal, timeout, review, and working without the model.

The browser never holds the API key. Calls go through a service you own, with authentication, rate limits, logging that removes personal data, and a defined response schema. The interface treats the model as an unreliable dependency, with timeouts and a manual fallback.

Standalone components, route configuration, functional providers, signals where they clarify the code, and RxJS where the domain is a data stream. Incremental migration from modules rather than a rewrite. Strict TypeScript and careful change detection. The goal is a system the team can maintain.

React projects at Hub Security and eToro. React on the client, Node.js and Express on the server, MongoDB when the data model fits. Shared UI as a small system. TypeScript preferred. MERN certification: 2022–2023.

Generation is a first draft. Prompts include your tokens, constraints, and examples so output follows your system. A person then reviews spacing, copy, focus order, and empty states. If the result looks generic, it is rewritten or discarded.

Yes. Design tokens (color, type, space, motion) stay shared in CSS. Each framework gets its own components: Angular standalone components and React function components, documented in Storybook. I do not share one runtime component across both stacks unless there is a clear reason.

It is part of the definition of done. Semantic HTML first, ARIA only when needed, visible focus, and keyboard paths. Automated checks (axe, Lighthouse) plus manual screen-reader review on critical journeys. Target: WCAG 2.1 AA.

Public targets are LCP, CLS, and INP. On Angular: signals or OnPush, trackBy, lazy routes, less work in templates. On React: memo where it helps, careful list rendering, and split bundles. Modern image formats and font-display swap.

Local UI state stays in the component. Server state uses a cache with invalidation (for example React Query on React, a clear data layer on Angular). NgRx when the domain is event-based or many features share one contract. Signals cover a large share of product UI without a global store.

Unit tests for logic that can fail quietly. Component tests for important interactions. A thin Playwright layer on journeys that affect revenue or trust. Lint, types, and accessibility smoke tests in CI. Pull requests stay small.

Yes. Hub Security and eToro included Angular systems and React projects. The hard part is the contract: routing, authentication, tokens, and shared UI rules. Independent deploys only help if design language and performance budgets are agreed. A clear boundary is better than a clever shared runtime that is hard to debug.

I work in English and Hebrew. Copy is stored outside components. CSS uses logical properties instead of left/right. Angular i18n or Transloco on Angular; a message catalog on React. Layout is tested in RTL, not only the translations.

I still implement the important paths. Around that: architecture the team can explain, reviews that teach, written decisions for choices that will be revisited, and small pull requests. Mentorship is pairing and checklists (accessibility, performance, security). At Hub Security the lead role is operational.

Model output is untrusted. It is encoded, never evaluated as code, and not allowed to instruct the interface to send data out. Tenant data stays out of logs. Standard OWASP controls still apply, plus prompt injection as a known abuse case. Features that cannot fail safely do not ship.

Production UI in the agreed stack (Angular or React), tokens and components, accessibility and performance work, and a short handover: what changed, what is deferred, and how to run it. Written notes or a recorded walkthrough when the team is distributed.

Fixed-scope sprints when the outcome is clear; a retainer when the product is still changing. I work with product, design, and backend. Based in Tel Aviv, available remotely, English and Hebrew. I typically reply within one business day.

A link or PDF of the current interface, the stack (Angular, React, backend, or a mix), whether AI is in scope, three priorities, and a date. I will reply with a short plan you can use even if you do not hire me.

Node.js and Express for APIs, Python where it fits, .NET/C# when that is the existing language. REST as the contract. PostgreSQL, MySQL, or MongoDB depending on the data. MEAN and MERN are both certified, working stacks.

Far enough that a merge can become a running system: GitHub, Jenkins, Cloud Run, Kubernetes, AWS, IIS. Secrets stay off the client. Backups are part of the design. Software architecture training: DevOps Experts College, 2024.