$ whoami
Victor Fialho — 20 · Brasília, Brazil 🇧🇷
AI Engineering @ IESB (one of Brazil's first bachelor's programs in the field)
Founder @ AoTomate Sistemas
I don't just study automation — I build it, ship it,
and debug it in production for real clients.
$ cat languages.txt
PT (native) · EN (fluent) · ES (advanced)
$ cat off_duty.txt
Jiu-Jitsu · philosophy of mind · history of mathematicsFlagship AoTomate system: an AI customer service agent built for a remote secretarial company, live in production. End-to-end ownership — architecture, deployment, monitoring, and production debugging.
┌─ STACK ──────────────────────────────────────────────────┐
│ N8N · Evolution API · Supabase · Groq (Llama 3.3 70B) │
│ Oracle Cloud (2 VMs) │
└──────────────────────────────────────────────────────────┘
- What it does: answers inbound WhatsApp messages 24/7 in seconds, routes conversations per tenant, escalates to humans only when needed
- Engineering highlights:
- Diagnosed and fixed a production-breaking change in Meta's
remoteJid→@lidGUID format that silently broke message routing in Evolution API - Multi-tenant message routing with hardened Supabase RLS policies
- Zero-marginal-cost inference pipeline — migrated from Claude API to Groq free tier
- Diagnosed and fixed a production-breaking change in Meta's
- Code: private (client production system) · architecture writeup available on request
STATUS: ● LIVE
Full-stack system built for a real organization (União Centro Oeste Brasileira). Complete order lifecycle — request → warehouse separation → shipment → delivery — with role-based access and photo evidence at each stage.
┌─ STACK ──────────────────────────────────────────────────┐
│ Next.js 15 · React 19 · TypeScript · Supabase │
│ Tailwind CSS 4 · shadcn/ui · Vercel CI/CD │
└──────────────────────────────────────────────────────────┘
- Features: multi-role auth (admin/operator), real-time inventory and shipment tracking, report generation, photo uploads
- Deploy: sistema-gestao-pedidos.vercel.app
STATUS: ● LIVE
NLP pipeline for sentiment analysis in Brazilian Portuguese — classical ML from scratch, no pre-trained models. Corpus preprocessing → TF-IDF → Logistic Regression, with focus on PT-BR text normalization, class imbalance, and model interpretability.
┌─ STACK ──────────────────────────────────────────────────┐
│ Python · scikit-learn · pandas · NLTK │
└──────────────────────────────────────────────────────────┘
The venture under VF-Labs. I build custom AI agents, WhatsApp automation pipelines, and management systems engineered around each client's actual operation — no templates, no bloat. First client live in production; building toward a SaaS product.
→ More at victorfiallho.github.io
$ cat roadmap.txt
[NOW ] Agentic AI — multi-agent orchestration beyond visual builders
[NOW ] Classical ML & NLP — scikit-learn, feature engineering, PT-BR text
[NOW ] AI Engineering degree — math foundations (calculus, linear algebra)
[NEXT] Cloud certification track — AWS AI PractitionerAI & Automation
Frontend
Backend & Data
Infra
DATA · MODELS · SYSTEMS · IDEAS · CODE
