Researcher.
Engineer.
Evaluator.
I build machine learning systems, then test them hard.
I’m Ines Varela, a research engineer in Lisbon. I take models from first experiment to production, and I build the evaluations that tell us whether they work.
I look for where models break
Leaked test data, and scores that rise while answers get worse. I log each failure and write a check for it.
How I work
One question, then many parallel runs
Test data leaks
Benchmarks drift into training sets. I check for overlap before I trust a score.
Metrics drift
A number can go up while answers get worse, so I read a sample of outputs every week.
Small evals first
A fifty-example eval you run today beats a perfect one you run next month.
Experience
Eight years from lab to product
Research and engineering roles in a university lab, a robotics start-up and an applied AI team.
2023 – now
Senior Research Engineer
Applied AI team, Example Labs
I lead evaluation for language model features used by about two million people.
Details
Built the team's evaluation harness: 40 task suites that run on every model change. Nightly eval reports cut the time from regression to fix from weeks to two days. I mentor four engineers on experiment design and code review.
2020 – 2023
Machine Learning Engineer
Acme Robotics
Trained and shipped perception models for warehouse robots.
Details
Moved training onto a shared pipeline so every experiment could be repeated. A relabelled test set cut false emergency stops by 31%.
2018 – 2020
Research Assistant
Vision Lab, Example University
Studied how image models fail when the data shifts.
Details
Co-authored two workshop papers on robustness benchmarks.
Ongoing
Teaching and talks
Independent
Guest lectures on evaluation and a yearly workshop for new ML engineers.
Details
Notes and exercises are published under a free licence.
Selected work
Three projects, start to finish
2024 – now
Open evaluation toolkit
Creator and maintainer
A small Python library for writing model evaluations as plain test files, with reports the whole team can read.
Read more →
2025
Retrieval that admits doubt
Lead author
A study of when a retrieval-augmented model should answer "I'm not sure", with a benchmark of 3,000 hard questions.
Read more →
2023 – 2024
Small models on small devices
Tech lead
Distilled a speech model to run offline on mid-range phones without losing the words people use most.
Read more →
Education
Computer science, then AI
2016 – 2018
MSc, Artificial Intelligence
Example University · Thesis on robustness under distribution shift
2013 – 2016
BSc, Computer Science
Example University · First-class honours
Award
Best workshop paper
Robustness workshop, 2025
Fellowship
Graduate research fellowship
Example Foundation, 2017
Talk
Evals your team will actually run
ML engineering meetup, 2024
Teaching
Guest lecturer, Applied Machine Learning
Example University, 2023 – now
Certification
Professional machine learning engineer
Cloud certification, 2022
Service
Reviewer
Two machine learning conferences, 2021 – now
How I work, in four lines
Contact
Let’s talk.
Open to research engineering roles and collaborations. Based in Lisbon, working with teams anywhere.