Software Engineer based in São Paulo 🇧🇷, currently building fintech and crypto products at Trace Finance with a globally distributed team.

I care about clean architecture, reusable systems, and shipping things people actually use. I like Anime, gaming, toy photography, baseball and NBA

• 🛒 Open-source e-commerce monorepo — a reusable foundation for new sellers to launch and scale their shops. Issues and PRs welcome.
• 💼 Day-to-day: fintech & crypto platforms at Trace Finance.
• 📚 Always exploring: edge databases (ElectricSQL), local-first apps, and DX tooling.

I'm deep into AI-assisted development — Claude Code is part of my daily workflow.

• Building custom agents, slash commands, and hooks to automate planning, reviewing, and scaffolding.
• Designing multi-agent code review (specialist reviewers for security, conventions, logic, tests) to raise the floor on PR quality.
• Sharing what works (and what doesn't) — prompt design, agent orchestration, and where AI still falls short in real codebases.

If you're building tooling around Claude Code, the Anthropic SDK, or agent workflows, I'd love to compare notes.

I'm build a AI project called, Lins Flow. That is a desktop app where you compose multi-step AI workflows on a node-based canvas, instead of writing prompts ad-hoc.

You drop blocks on a board — triggers, pipelines, artifacts, repositories, agent pools, context — connect them with edges, and each pipeline runs Claude Code locally with the skills, agents, and context you wired in. Outputs flow downstream as typed artifacts (subtasks, code, PRs, reviews), so a real feature can be expressed as Plan → Execute → Review → PR, parallelized across subtasks, and re-run from any failed step.

Under the hood:

• ⚛️ Electron + React 19 renderern XY Flow
• ⚡ Fastify + PostgreSQL + ElectricSQL backend with WAL-streamed local-first sync
• 🔌 TanStack DB + Socket.IO for ptimistic writes and ACK-requiredmutations
• 🧠 Local Claude CLI execution via node-pty, with an MCP server bridging tool calls back into the app

It's a real engineering exercise in local-first architecture, multi-user realtime sync, and DAG-based execution — the kind of system where a single SSOT decision in the wrong place ripples for weeks.

What I reach for when starting or scaling a frontend project today:

• Languages: TypeScript, JavaScript, Kotlin
• Frameworks: Next.js, React, React Native, TanStack Start, Electron
• State & data: Zustand, TanStack
• Forms & validation: React Hook Form, Zod, Yup
• Styling: Tailwind CSS, styled-components, PandaCss, StitchsJS, Design Systems
• i18n: next-i18next
• Auth: NextAuth.js, JWT, OAuth
• Testing: Jest, Vitest, React Testing Library, Playwright
• Quality & DX: ESLint, Prettier, Biome, Husky
• Monorepo & build: pnpm workspaces, Turborepo, Vite, esbuild
• Backend (when needed): Node.js, NestJS, Express, Fastify, GraphQL, Ktor
• Databases: PostgreSQL, SQLite, ElectricSQL, Drizzle ORM, Prisma
• CI/CD: GitHub Actions, Vercel, Railway, Docker
• Observability: Sentry, Datadog
• AI tooling: Claude Code, MCP servers, agent orchestration

English CV

• LinkedIn: viniz-arruda
• X / Twitter: @viniz_arruda
• Email: viniarruda.souza@gmail.com
• Stack Overflow: profile