You don't want another contract that ends at the notebook. You want the pager, the promotion lane and the authority to say a model isn't ready. This one is based in Brazil, 40 hours a week on your own invoice, and everything after handoff is yours.
Location: Remote from Brazil — you must be based in Brazil. 40 hours per week, Monday to Friday, with daily overlap into the Berlin working day.
About us
CarOnSale is the AI-powered platform for B2B used car trading in Europe. Over 40,000 buyers from more than 20 countries trade on our platform — and 85% of inventory is exclusive to us. We connect software, pricing intelligence, logistics and financing in one layer — as the operating system for an entire industry.
One Platform. One Profit Engine.
The platform you build in
Our machine learning runs on one shared, central platform — not a separate pipeline per model. Five canonical stages: data extraction, validation, transformation, training and evaluation. A Snowflake data warehouse feeds a SageMaker managed feature store, and models reach production through governed CI/CD promotion lanes on Terraform-managed AWS infrastructure. Four models serve production today. Fifteen to twenty by mid-2027. Your job is to build inside that platform and make it stronger, so the next model costs less to ship than the last one.
Your responsibilities
- You own models from handoff through to production: packaging, deployment, monitoring, and the decision on whether a model is ready to serve
- You keep production models reliable — drift detection, performance monitoring, alerting and incident response when something moves
- You own the serving and inference path: fitted pipeline artifacts, inference entry points, monitoring hooks and feature-store parity
- You review model design and evaluation methodology before anything ships, and catch data leakage, backward-window errors and weak evaluation during development, while they are still cheap to fix
- You extend the shared platform so it stays useful for every model, without project-specific logic leaking into shared code
- You set the engineering standards the platform runs on as it scales across the organisation
What you bring
- 2+ years in production machine learning engineering, with real ownership of models after handoff — not only training them
- Strong Python: typed, tested, production-grade code, and you review the work of others
- Enough machine learning depth to challenge a pipeline on problem framing, feature engineering, model selection and evaluation methodology
- Hands-on experience with a managed ML platform — SageMaker, Vertex AI, Databricks or Azure ML — plus feature stores, CI/CD for machine learning, AWS and Terraform
- An AI-native way of working: you use tools like Claude, ChatGPT or Copilot actively in your daily work
- English at C1 level, written and spoken. German is not required — we work in English
- You are based in Brazil and invoice through your own company. We work directly with you, not through intermediary or umbrella services
Nice to have
- Snowflake and dbt — you can pick both up here
- Experience mentoring colleagues or reviewing their work
- Comfort operating where the answer is not defined yet
What to expect from us
- A full-time engagement: 40 hours per week, Monday to Friday, invoiced monthly against your own company
- You are treated like a full member of the team — standups, bi-weekly sprints, and all company communication
- Fully remote from anywhere in Brazil
- An English-speaking engineering team with short decision paths
- Direct ownership of models serving a live product, not a proof of concept
- Structured onboarding with a buddy from the team
Apply now — your CV is enough.
