Available for projects

~/ai-engineer

I build AI agents that run in production, not demos.

WhatsApp agents that answer, qualify and book appointments for 40+ businesses.

WhatsApp · AI agentagent execution

● status · systems in production

Not portfolio projects. Systems clients rely on every day.

15k messages a day · 40+ businesses served · ~5k appointments booked

Real production numbers · updated Sep 2026

cases · 04

What I’ve shipped to production

01 · featured

Multi-tenant AI agent platform on WhatsApp

Agents that answer, qualify and book for 40+ businesses. Migrated from n8n to a code-first Kestra architecture with a custom agentic loop and GitOps deploys.

  • 40+ businesses
  • 90% without humans
  • 3 months, zero downtime

kestra · python · supabase · whatsapp · openrouter

Read the case →
Architecture: WhatsApp sends messages to Kestra, which runs the receiver and a 12-tool agentic loop, integrated with LLMs, Supabase, Chatwoot and calendars.whatsappcloud api · wahakestra · 3 namespaceswa-message-receiveragent-loop12 tools · typed i/ollmopenrouter · geminisupabasepostgres · rlschatwoothuman handoffcalendarsgoogle · amigo · gdsArchitecture: WhatsApp sends messages to Kestra, which runs the receiver and a 12-tool agentic loop, integrated with LLMs, Supabase, Chatwoot and calendars.whatsappcloud api · wahakestra · 3 namespaceswa-message-receiveragent-loop12 tools · typed i/ollmopenrouter · geminisupabasepostgres · rlschatwoothuman handoffcalendarsgoogle · amigo · gds
  1. 02AI ticketing hub · SaniiAI reads ~140 WhatsApp groups and turns absences and complaints into tickets.n8n · gemini · airtable~99 groups
  2. 03Migration to the official WhatsApp APIProduction agents moved from unofficial APIs to Meta’s Cloud API.graph api · cloud api v25 · n8nreusable runbook
  3. 04AI SDR for a travel agencyQualifies international flight leads and hands a ready brief to the sales rep.n8n · supabase · redisofficial cloud api

services

How I can help your operation

AI agents on WhatsApp

They answer, qualify, book and follow up around the clock, and hand off to your team at the right moment.

sdr · scheduling · follow-up · human handoff

what changes when the agent comes in

  • ~30 min60 sfirst response time
  • 100% human90%of conversations solved by the agent
  • no-shows−50%with automated reminders

how it works

Four steps. Then the agent works and I keep watch.

  1. 01Call30 minutes to understand your operation and where support is losing sales.
  2. 02DiagnosisI map flows and integrations and lock scope, timeline and price.
  3. 03BuildWeekly deliveries, tested with real conversations before going live.
  4. 04OperationMonitoring, tuning and reports. I don’t ship and disappear.
Marco Fabian smiling, in a white shirt, holding a coffee mug

about

Hi, I’m Marco.

I’ve been writing software professionally since 2022. I started with Node.js APIs, moved through ERP systems and integrations, and since 2025 I’ve been building AI agents for real businesses. I’m studying Computer Science at UFMG and graduated in Full Stack Development from Trybe.

The first agents were built in n8n. As volume grew, I moved everything to a code-first architecture: Kestra, Python, Git, CI. Today I lead the AI implementation team at Clavia, where our agents serve 40+ businesses on WhatsApp. I joined as an engineer and was promoted to lead after being named Employee of the Month.

Belo Horizonte · Clavia · High.AI · Somedi.AI · NetZero · UFMG · Trybe

stack

ai & agents
agentic loops · tool calling · rag · versioned prompts · observability
llms
openrouter · claude · gemini · gpt
orchestration
kestra · n8n
languages
python · typescript · javascript · sql
data
supabase · postgres (rls) · redis · mysql · mongodb
integrations
whatsapp cloud api · waha · uazapi · chatwoot · google apis · feegow · gestãods · amigo
infra & testing
docker · github actions · hetzner · jest · pytest

Available for projects

Ready to put an agent into production?

Tell me about your operation. I reply within one business day.