Gustavo Scherer
ML Engineer
open to ml engineering and backend roles
i work on production python backends — apis, event-driven pipelines, and the infrastructure around them — and i'm now focused on bringing ml models into production through tracking, versioning, and serving. i'm pursuing a postgraduate degree in ml engineering at fiap and live in foz do iguaçu, brazil. outside work, it's counter-strike and tinkering with my own server and homelab.
$ highlights
- python in production at an ad-tech company since jan 2025: event pipelines, message queues, ci/cd on kubernetes
- ml pipelines end to end: dvc, mlflow model registry, dockerized fastapi serving — roc auc 0.92 on purchase intent
- built and run a web map game: 62k+ houses placed, 100k+ visits
- i run my own production servers: nginx and caddy, docker compose, jenkins and github actions, postgres
$ experience
2 roles · since 2022
-
Backend Engineer
– now
JoinAds1 yr 10 mos
- migrated background jobs from RQ/Redis to RabbitMQ with delayed messages, and added an outbox for reliable event delivery, with its own worker and metrics
- FastAPI services and integrations with the Google Ad Manager, Google Ads and Meta APIs
- shipping through Docker, GitHub Actions, GHCR and ArgoCD on K3s; structured logging and Prometheus metrics in production
-
Freelance Software Developer
– now
self-employed4 yrs 10 mos
- end-to-end ownership for small-business clients, from requirements to deploy
- a management system, a Next.js + Prisma e-commerce store and a WhatsApp customer-service copilot running in production, plus landing pages with Pix checkout
- workflow automation and third-party integrations that replaced manual operations
$ education
Postgraduate, Machine Learning Engineering
–
FIAP · model lifecycle, inference APIs, MLflow, DVC, Docker
Technology Degree, Systems Analysis and Development
–
IFPR – Federal Institute of Paraná · final project: LingoLearn
$ projects
purchase-intention
view project →predicts whether an e-commerce session ends in a purchase. random forest with class balancing: roc auc 0.92, recall 0.81 on the test set. reproducible dvc pipeline, mlflow tracking and model registry (@champion alias), multi-stage non-root docker image, fastapi /predict. my part: repo structure, inference api, dockerfile and api tests. fiap tech challenge, team of 5.
python · scikit-learn · dvc · mlflow · docker · fastapi · poetry
churn-prediction
view project →telco churn on 7,043 customers. picked logistic regression over a random forest with slightly higher auc (0.849 vs 0.852) for better recall and f1 and cheaper serving: recall 0.78, 5-fold cv f1 0.640 ± 0.021. leakage columns removed, 57 pytest tests. my part: fastapi inference api and repo ownership. fiap tech challenge, team of 5.
python · scikit-learn · fastapi · pytest · uv
tijolo a tijolo
visit site →a web map game where players place houses city by city: 62k+ houses and 100k+ visits, grown through instagram reels (one reel passed 50k organic views). runs on my own vps with jenkins ci and postgres; i handle the ops, from container memory limits to database load.
javascript · node.js · postgres · jenkins · docker
concord
private repoself-hosted private chat with text, voice/video and screen sharing through a mediasoup sfu with two-layer simulcast. react frontend and fastify services for auth, media and chat; postgres behind caddy; ci/cd with github actions and docker compose, with rollback and automated postgres backups. in production on my own vps.
typescript · react · fastify · mediasoup · webrtc · postgres · caddy · docker
lingolearn
view project →learn languages by reading: text library, leveled vocabulary, progress tracking, multiple languages and jwt auth. my undergraduate final project.
sveltekit · fastapi · python · sqlite
$ skills
The tools I work with, from backend services to model serving.
$ contact
you can find me at: