Arthur Kellermann
Engineer, builder at heart, and a firm believer that the best products still get built by one person who cares too much.
I redirected my career to do one thing: help startups go from zero to a shipped product. I work with founders around the world who move fast and care about what they're building. If that sounds like you, don't hesitate to reach out. I'd love to talk.
I'm most alive at zero to one: the stretch where a product doesn't exist yet and every decision still matters.
Experience
The roles and the moments that actually shaped how I build.
Senior Software Engineer
Working directly with early-stage founders in healthcare, fintech and e-commerce to ship 0→1 MVPs and production-grade AI features, end-to-end, from first architecture decision to what's still running a year later.
Dell Technologies · Intern → Software Engineer II
Joined as an intern shipping tickets on an internal enterprise platform; earned my way to leading its development: not just the engineering, but the technical decisions and adoption strategy across the organization. Owned the full SDLC and served as the main technical point of contact for bringing AI capabilities into business workflows across the platform.
1st place · DTX Hackathon
Theme: transforming operations with AI-driven automation. Built and deployed a Python API on Llama's LLM, integrated into production tooling with structured prompt engineering, in the event window, not after it.
B.Sc. Computer Science · UNISINOS
Universidade do Vale do Rio dos Sinos, São Leopoldo, Brazil.
Selected work
Most of what I build is under NDA. But I always carve out time for my own projects, built on whatever's newest, to stay sharp and current with the market.
An agentic AI assistant
- Built as an explicit LangGraph state machine: intent routing into specialized branches, a deterministic multi-step subgraph for form recommendation, and Postgres-checkpointed conversation state enabling human-in-the-loop approval that survives across sessions.
- Designed a generative UI: graph nodes emit versioned semantic intent events over SSE, resolved by an Angular client into design-system components, kept rendering framework-agnostic rather than pushing server-rendered components.
- Exposed every agent capability through MCP so new systems integrate at the boundary instead of inside the graph, and kept user identity out of the model's tool vocabulary, enforced server-side instead.
- Gated releases on an LLM evaluation suite running in CI.
Stack: Python · FastAPI · LangGraph · MCP · PostgreSQL · Langfuse · Angular
Sandbar Rally
A first-person beach volleyball game, built purely for the fun of it: passing, setting, spiking and blocking against a CPU or a friend, single HTML page, no build step.
I use side projects like this one to stay current with what's actually shippable, not just what's announced: this was a stress test of Claude Fable 5.1 running as a multi-agent swarm, orchestrated through Orca as the agent development environment: planner, builder and reviewer agents coordinating against one shared codebase in real time.
Stack: Three.js · JavaScript · Claude Fable 5.1 · Orca (multi-agent ADE)
Stack
What I reach for to get a product from idea to production, and keep it there.
- Angular
- React
- React Native
- TypeScript
- Python
- Node.js
- LangGraph
- Agents & MCP
- RAG, embeddings
- Langfuse
- Guardrails
- Evals in CI
- AWS
- Docker
- Kubernetes
- CI/CD
- PostgreSQL
Let's
talk.
Whether you're building something, hiring, or just want to trade notes on shipping fast, I'd love to hear from you.