Software Engineering
TypeScript and Node.js end to end: React and Next.js frontends, Fastify/NestJS APIs, PostgreSQL, monorepos, CI/CD on Railway and Cloudflare. Typed, tested, deployed.
Open to remote roles · US/EU time zones
AI agents and automations that run real businesses.
I'm Paulo Chaves, a software engineer from São Paulo, Brazil. I design and ship LLM-powered agents, n8n automation systems, and the web platforms around them — in production, with tests, for real customers.
Senior Software Engineer · Applied AI & Automation
Capabilities
What I do
✦Engineering that connects interfaces, systems, and operational decisions.
Applied AI
LLM agents in production: OpenAI, Anthropic and Gemini models, RAG with pgvector, tool-calling agents, evaluation suites, and guardrails against hallucination. AI that answers correctly — or hands off to a human.
Automation (n8n)
WhatsApp-first automation for e-commerce and service businesses: n8n at the core, Z-API and Evolution API for WhatsApp, ERP and store integrations. Governed like software: versioned, tested, observable.
Public kits
Open source
✦Public kits for turning automation into a repeatable engineering practice.
One-click n8n deployment on your own Railway account, plus a governance layer to operate it with Claude Code.
Versioned workflows, automated tests, and safe promotion between environments. MIT licensed.
- Automated tests in CI
- 304
A complete WhatsApp commerce agent for n8n: product search, shipping quotes and order tracking (Nuvemshop + Correios + Z-API), plus a Next.js management panel with human takeover.
44 sanitized production workflows, test-first.
The case behind it: built for a Brazilian jiu-jitsu kimono e-commerce. The agent served 333 customers, resolved 63% of conversations without human help, and was attributed R$27.8K in sales. The first version took 45 days to build. Rebuilt with the governance kit, the second version took 7.
- Production workflows
- 44
- Conversations resolved
- 63%
- Build cycle
- 45 → 7 days
Case studies
Selected work
✦Client names are omitted on purpose. I can walk through any of these in a call.
AI Products
Legal AI for agribusiness attorneys
An assistant that reads full contracts and rural land records — including scanned pages, via vision OCR — and answers with page-level citations.
1M-token context window, no RAG by design: the whole case fits in the window.
- 1M tokens
- Context window
AI speech-coaching platform
Users record speeches and get a standardized 18-point analysis, then practice against AI conversation avatars.
Speech-to-text, multi-voice TTS, LLM scoring, subscriptions with a credit wallet.
- 18
- Analysis points
Vertical CRM for aesthetics clinics
Clinic management, sales, scheduling and a native WhatsApp inbox in one platform.
NestJS + React monorepo, 1,100+ automated tests, designed to scale to a thousand clinics.
- 1,100+
- Automated tests
White-label AI agent platform
Chat agents with RAG knowledge bases (pgvector), role-based access and analytics.
Deployed as themed verticals for different niches, from digital marketing operations to real-estate development.
- pgvector
- Knowledge layer
AI content studio for beauty professionals
Eleven specialized AI agents — reels, stories, editorial calendar, pricing — in a chat interface with live artifacts.
- 11
- Specialized agents
WhatsApp group intelligence
A SaaS that monitors WhatsApp groups and qualifies leads through a gated funnel.
Fastify, Prisma, BullMQ, Evolution API.
- In development
- Status
Client Work
AI assistant for an accounting firm
A WhatsApp agent integrated with the firm's ERP, serving ~36 business clients.
Built with a target of resolving 80% of requests without a human.
- 80%
- Resolution target
Operations hub for a digital marketing group
Consolidated six internal tools into one app: single sign-on, role-based access, and a per-action audit trail.
~48 routes mapped across the operation.
- ~48
- Mapped routes
Payment studio for a US at-home wellness company
Replaced an n8n form flow with a React app that generates payment links and branded PDF payment requests.
- React app
- Delivery
Tooling
Stack
✦Tools selected to ship, observe, and sustain production software.
Languages
- TypeScript
- Node.js
- Python
- SQL
Frontend
- React
- Next.js
- React Router
- Vite
- Tailwind CSS
Backend
- Fastify
- NestJS
- Express
- Prisma
- Drizzle
- PostgreSQL
- Redis
- BullMQ
AI
- OpenAI
- Anthropic Claude
- Gemini
- RAG (pgvector)
- Tool calling
- Eval suites
Automation
- n8n
- Z-API
- Evolution API
- WhatsApp Business
Infrastructure
- Railway
- Cloudflare Workers
- Docker
- Supabase
- GitHub Actions
- pnpm monorepos
Built for the work after launch.
I'm from São Paulo and lived in London before coming back to Brazil to build with AI. Over the last years I've shipped AI products and automation systems for companies in e-commerce, accounting, legal and healthcare — always with the same principle: systems have to keep working after launch. That means versioning, tests, observability, and a clear handoff between AI and humans. I work in English and Portuguese.
Let's talk.
I'm open to senior software engineering roles — remote, US/EU time zones. LinkedIn and GitHub are the best ways to start a conversation.