Matheus Gonçalves ML & AI Engineer

Português

Greater Porto Alegre, Brazil · Open to remote roles worldwide · Open to offers

View resume Contact Projects

real embeddings · 2D UMAP · hover it

About

Product Analyst at Aepio, where I build the data systems behind telemetry-enabled gas meters.
I started at the calibration bench and moved into code: from the lab to product.
Now I'm in a graduate program in ML Engineering (FIAP), taking that data from recording to prediction.
Outside of code: football, games, barbecue and cars.

By the numbers

15k+meters/month flowing through the test benches I integrated
30k+meters in the field sending telemetry
24 h → real timetest data availability (from daily batch to real time)
4roles in under 3 years, from lab assistant to product analyst

Projects

6 test benches
Local queue + retry
Ingestion API
Single data contract
Approval check

In production

Aepio · test benches

Integrated 6 meter test benches with the central server, with automatic retry and a single data contract.

Python
REST APIs
MySQL
SQL Server

How I built it →: Data that couldn't be lost on a network that keeps going down

Launching

Sinalo

Collects public posts and ranks them with embeddings and a score, delivering only likely customers along with the reason.

Python
Scraping
Embeddings
Classification score

How I built it →: Finding leads in the noise of social media Website ↗: Sinalo

Launching

Gotem

Scans search engines, websites and Telegram for copies of digital products, comparing both text and images.

Python 3.12
FastAPI
Postgres + RLS
Task queue

How I built it →: Hunting pirated copies across the web Website ↗: Gotem

Launching

Embalo

An affiliate program for SaaS that tracks commissions automatically, straight from payments.

TypeScript
Next.js
Supabase
Stripe App

How I built it →: Affiliate commissions tracked without spreadsheets Website ↗: Embalo

Experience

Product Analyst · Technology & Development · Aepio

03/2026–present

Lab Assistant → Lab Operator → Senior Lab Assistant → Product Analyst, since Jan 2024

Built the Python REST API that connects 6 test benches to the central server. 15k+ meters/month flow through them, with approval checked automatically.
Moved test data from a daily batch to real time, eliminating 1 h/day of manual extraction by the Quality team.
Developed systems and telemetry analysis for a fleet of 30k+ meters in the field.
Designed the production and telemetry databases (MySQL, SQL Server) and built dashboards, automated test validators and field-support apps.
Integrated LLMs (Claude, Codex) into the daily engineering workflow.

Education

Graduate program in Machine Learning Engineering · FIAP08/2026 – 07/2027 (in progress)
Systems Analysis and Development (tech degree) · Centro Universitário CesucaCompleted 12/2025

Lab

ML experiments running right in the browser.

Drawing

The game at the top: a convolutional network trained in PyTorch on 240,000 drawings. It gets 93.1% of 30,000 test drawings right across 30 categories and runs in the browser with no library. ↑

Data: Quick, Draw! Dataset, Google, CC BY 4.0

Transformer attention

Hover a word to see where DistilBERT is looking.

I started at the calibration bench and now I build the systems it runs on.

real attention · distilbert · layer 5 · head 5 · per-word weights, renormalized

Latent space

The cloud at the top: 2,400 sentences turned into embeddings (multilingual MiniLM) and projected to 2D with UMAP. ↑

Stack

ML & AI: LLM
Embeddings
Machine Learning
PyTorch
Neural networks
CNN
Sentence Transformers
Data: Ingestion pipelines
IoT telemetry
SQL
MySQL
SQL Server
Postgres
Data modeling
Backend: Python
FastAPI
SQLAlchemy
Alembic
Task queues
REST APIs
Frontend: TypeScript
React
Next.js
Tailwind
Infra & quality: Git
Docker
Fly.io
Cloudflare
Supabase (RLS)
pytest
Vitest
ruff

Contact

Open to ML Engineer and AI Engineer roles, remote or hybrid.

Advanced English