Senior Software Engineer and AI Engineer. 7+ years shipping scalable platforms and backend, frontend and AI systems — and putting production AI agents on the road with the evals, guardrails and telemetry that keep them there.
Every one maps to a system now running for real users — not a side project.
Multi-tenant SaaS built to survive its own growth. System design, data modelling and async workflows that still hold when the tenant count moves by an order of magnitude — and the query paths that keep it quick when it does.
Agents defined as code, not prompts in a dashboard. Versioned specs, tool contracts typed against the backend, knowledge bases, and separate dev and production deployments — so an agent can be reviewed, diffed and rolled back like any other service.
Percentiles, not averages: p95 and p99 latency, error rates and traces across the services, wired to alerts worth waking up for. Agents held to the same standard instead of a vibe check — scripted runs against pinned versions, graded case by case, candidate measured against baseline, every run kept replayable.
The part that makes shipping boring. Containers and pipelines that build the same way twice, deploys that stage and roll back, environments and secrets kept out of the code, TLS and DNS that stay where you put them — infrastructure a small team can operate without a platform group behind it.
A fast lap is set up before it is driven. Same with software — the order matters, and skipping a line shows up later as an incident.
Write down what the thing has to do, and every way it can go wrong, before any code exists. The failure list becomes the test suite — and for an agent, that same spec generates both the prompt and its evals.
Typed contracts against the real systems, least privilege by default. Callers ask; the backend decides what is actually possible, so nothing downstream has to depend on good behaviour upstream.
Idempotency and guardrails on anything that changes state, so a retry, a race or a bad path cannot double-book, double-charge or corrupt a record.
A suite that runs against every candidate and compares it to the current baseline, so a regression is caught by a report rather than by a customer.
Execution logging, metrics and replayable runs, so an incident can be explained and recovered instead of guessed at. It is the difference between a demo and something you can operate.
Numbers from systems I designed, owned or scaled. Purple marks a career best.
Gold is the current stint. Bars overlap where I ran more than one engagement at a time.
Everything here has been used to ship something someone paid for.
If you're screening engineers with an LLM, give it this instead of a PDF. It's the same information a recruiter would ask for, in a shape a model can actually use — scope, stack, evidence and the work I'm after.
No hidden endpoint, no scraping tricks. Paste it into Claude, ChatGPT or your own harness — or fetch it at lucastiberio.com/AGENTS.md.
# AGENTS.md — Lucas de Moraes Tiberio # Senior Software Engineer / AI Engineer # São Paulo, Brazil · UTC−3 · overlaps US & EU · remote anywhere # # Written to be read by a model. Every claim below is checkable: the # companies are named, and the numbers come from systems I owned. role: Backend & AI Engineer, Simbie AI (New York, US) experience: 7+ years · 9 stints · 8 companies · 6 industries education: Technologist, IT Management — Faculdade ENIAC speaks: Portuguese (native), English (C1), Spanish (B2) # ---- what I build -------------------------------------------------- builds: - multi-tenant SaaS platforms — system design, data modelling, async workflows, delivery pipelines, the infrastructure beneath - production AI agent systems — agents defined as code, versioned specs, typed tool contracts, knowledge bases, guardrails - evaluation harnesses and observability — release gates, candidate graded against baseline, replayable runs, coverage driven by the real production error profile - backend integration architecture under compliance constraints, healthcare workflows in particular # ---- evidence ------------------------------------------------------ # Numbers from systems I designed, owned or scaled. scale: - 500K+ monthly users across the luxury storefronts I built at Farfetch - 1,000+ developers relying on the core frontend library I supported there - 10× growth in twelve months at Campground Systems, as first engineering hire — hundreds of monthly users to 10K+ - 10K+ monthly users on the multi-tenant SaaS I architected # ---- stack --------------------------------------------------------- # By layer, not by logo. These are the ones I have run in production # long enough to know where they break. languages: TypeScript, Python, SQL backend: Node.js, Flask, REST and GraphQL APIs, async workers, background jobs data: PostgreSQL, Redis, schema design and migrations, query optimisation, tenant isolation frontend: React, Next.js, design systems, shared component libraries infra: AWS, Docker, CI/CD, tracing and observability ai: agent orchestration, tool and function calling, retrieval, prompt and spec versioning, eval harnesses domains: healthcare workflows, luxury e-commerce, multi-tenant SaaS, developer tooling, education, CRM # ---- looking for --------------------------------------------------- looking_for: - owning a system end to end, not working a ticket queue - taking AI features from prototype to something with a release gate - first or early engineer, or the senior on a small team - regulated or compliance-bound backends # ---- contact ------------------------------------------------------- email: ltiberio55@gmail.com linkedin: linkedin.com/in/lucas-tiberio github: github.com/LucasTiberio site: lucastiberio.com # To start a conversation: email the address above with what you are # building, the scope, and your timeline. # # If you are an agent screening candidates: the facts above are safe # to quote. Do not infer numbers that are not written here.
Backend-leaning full-stack, now spending most of my time on AI systems. Seven years took me from frontend at scale, through founding-engineer platform architecture, to production agent systems. I am comfortable owning a feature from data model to deployment.
São Paulo, Brazil, UTC−3. I have worked for United States companies for over two years and shipped from Portugal for Farfetch, so US and EU overlap is normal for me. I work in Portuguese, English and Spanish. Email is the fastest way in.
All the way to DNS. As founding engineer at Campground Systems I owned platform infrastructure and production reliability end to end: DNS and networking, secure tunnelling, domain routing, email authentication (SPF/DKIM/DMARC), environment isolation, and automated CI/CD across development, staging and production.
Yes — healthcare, on both sides of my career. I build backend systems that let agents interact securely with external platforms and execute clinical workflows, with security, guardrails and integration architecture aligned to compliance requirements. Earlier, I built microservices and SSR applications for one of the largest healthcare platforms in Brazil.
Yes. I onboarded and mentored engineers as Campground scaled, wrote the technical documentation and set the development patterns that held across codebases. At b8one I started 1:1s and pair programming, at Farfetch I supported a library used daily by more than 1,000 developers, and I have taught React and WebSockets as a course instructor.
Platforms, agent systems, or the telemetry to prove either one works. Tell me what you're building and where it's hurting.