About Me

I am a computer scientist who builds and leads applied AI systems, with a focus on LLM agents and the reliability of data-driven systems. I am the founding Scientific Director of Instituto Curvelo, a non-profit R&D institute, and a postdoctoral researcher at the Federal University of São Paulo, where I develop Mensura: a statically typed language for data handling that turns data leakage into a compile-time error, with its core algebra formally verified in Lean 4. I teach and advise graduate students at ITA and Unifesp, and I wrote a book on data science projects focused on validation and data leakage.
Above all, I am a Christian. My faith in Christ comes first and shapes how I live, work, teach, and lead. I am a happy husband and a proud father of two. In my free time: programming, science fiction (books and movies), and lots of coffee ☕.
The fear of the Lord is the beginning of knowledge; fools despise wisdom and instruction. (Proverbs 1:7, ESV)
Industry R&D
Most of my current work is applied R&D with companies, usually under non-disclosure agreements, so it does not appear in my publication list. I design and build LLM-agent systems and lead the teams that ship them: agent architectures that act through tools (MCP) under scoped access and human approval, safeguards around agent actions, evaluation harnesses that test agents against simulated users with LLM judges in continuous integration, and retrieval-augmented generation over enterprise data. I still write a good part of the code myself, mostly in Go and Python. At Instituto Curvelo the same approach extends to computer vision, embedded systems and IoT, and industrial automation.
Research and Engineering Themes
- Reliable and verifiable LLM systems. Evaluation metrics for language models (ICICC, 2025), theses on active testing to reduce human annotation in LLM evaluation and on LLMs for automated program repair, and ongoing master’s research on active testing and evaluation of LLMs.
- Type systems and formal verification for data science. Mensura, where data leakage is a compile-time error, built on a split-invariant algebra verified in Lean 4; the same concern for validation drives both of my books.
- Language models for software and security. Vulnerability detection and interpretability with language models (LADC, 2025).
- Graph-based retrieval and graph machine learning. A higher-order topological framework for RAG (BRACIS, 2026), and complex networks for prediction problems (Physica A, 2026).
- Machine learning for aerospace and the environment. Wildfire risk prediction from drone, satellite, and ground-sensor data (IEEE Access, 2024), GNSS-free image-based navigation with the Brazilian Air Force, IEAv, and Embraer (Sensors, 2024), and DroneComp.
Highlights
- Founded Instituto Curvelo in 2025 (legal structure, R&D model, team, and project portfolio) and lead its technical direction.
- Raised and coordinated R&D funding at ITA from Finep, the Brazilian Air Force, Embraer, Apex Brasil, CAPES, and CNPq.
- Co-designed, taught the first class of, and coordinated ITA’s Data Science Specialization Course (about 110 students in three cohorts); coordinated undergraduate research for the whole of ITA (2021-2024).
- Advised 5 master’s and 4 PhD students, plus 15 specialization and 14 undergraduate research students.
- Author of Projeto de Ciência de Dados (LTC/GEN, 2026) and Data Science Project (free on Leanpub); a graduate textbook on generative AI is under development for LTC.
- BSc and PhD in Computer Science from the University of São Paulo, both with honors; PhD thesis among the three best of the year.
Selected Publications
- Unifying graph-based retrieval: a higher-order topological framework for RAG. BRACIS, 2026. DOI
- Source code vulnerability detection and interpretability with language models. LADC, 2025. DOI
- Graph machine learning for flight delay prediction due to holding maneuver. Physica A, 2026. DOI
- Drone-based AI system for wildfire monitoring and risk prediction. IEEE Access, 2024. DOI
- Network unfolding map by vertex-edge dynamics modeling. IEEE Transactions on Neural Networks and Learning Systems, 2018. DOI
Full list here.
Experience and Education
- 2025-present: Founding Scientific Director, Instituto Curvelo
- 2025-present: Postdoctoral Researcher, Unifesp (supervisor: Marcos G. Quiles)
- 2020-present: Affiliate Professor, Graduate Program in Operations Research ITA/Unifesp; teaches Machine Learning (PO-233) and Data Science Project (PO-235)
- 2018-2024: Assistant Professor, Aeronautics Institute of Technology (ITA)
- 2014-2018: PhD in Computer Science (Hons.), University of São Paulo (advisor: Liang Zhao); visiting scholar at Arizona State University in 2016 (supervisor: Ying-Cheng Lai)
- 2010-2014: BSc in Computer Science (Hons.), University of São Paulo
For Grad Students (and Undergrad Researchers)
- Pro Git Book: I will always assume you know Git.
- The Not So Short Introduction to LaTeX: and LaTeX too.
- Awesome research plan template by Prof. Loubach: extra resources here (just ignore the Emacs section :P Vim rules supreme!)
- You can use any programming language you are more comfortable with. Just make sure that your research is reproducible.
Links
GitHub · LinkedIn · Google Scholar · ORCID · Lattes · Scopus · Web of Science · ResearchGate · Linktree
Contact Information
- Email: filipe.verri@institutocurvelo.org.br
- Phone: +55 (12) 99723-9684