Marina Ramalhete Masid

Senior Data Scientist & Machine Learning Engineer Master's Student in Computer Science

Brazil

I work on generative AI, knowledge graphs, and production NLP systems, across the full ML lifecycle from dataset curation to deployment. My work centers on LLMs, BERT-based architectures, embeddings, and metadata extraction that turn unstructured text into queryable knowledge. I have built named entity recognition and entity linking systems, information retrieval solutions, and LLM evaluation pipelines.

Experience

Jusbrasil –present

Remote · jusbrasil.com.br

Senior Data Scientist | Machine Learning Engineer –present

Scale the company's use of generative AI, taking state-of-the-art NLP into shipped products. Work on the architecture and productization of LLMs for classification and unstructured data extraction, and build the evaluation pipelines that make model behaviour predictable, reliable, and safe before it reaches users.

Data Scientist | Machine Learning Engineer –

  • Developed and evaluated LLM-based systems for classification and entity extraction, focused on performance metrics and evaluation pipelines.
  • Led the end-to-end lifecycle of NER models (deep learning, BERT), from specialized dataset curation with Prodigy and spaCy through production deployment in the legal domain.
  • Implemented entity linking by integrating extraction models with knowledge graphs, structuring unstructured data to generate contextual insights.
  • Applied text similarity techniques and classical supervised models to process large-scale legal datasets.

Globo –

Remote · grupoglobo.globo.com

Machine Learning | Software Engineer –

  • Designed, built, and maintained the team's semantic API and platform.
  • Developed and maintained knowledge graphs for Globo products, including metrics for semantic product usage.
  • Owned architecture, implementation, deployment, and maintenance of the team's services.
  • Implemented and maintained the team's machine learning models.

Machine Learning Engineer Intern –

  • Studied semantics and knowledge graphs for Globo products.
  • Software maintenance, and development and monitoring of metrics on semantic product usage.

Universidade Federal Fluminense –

Volta Redonda, Rio de Janeiro, Brazil · uff.br

Undergraduate Student Researcher –

Research spanning artificial intelligence, economics, big data, industrial organization, Industry 4.0, information technology, streaming media, and price discrimination. The work produced my undergraduate thesis, Streamings in Brazil: characterization and considerations of algorithms.

Teaching Assistant, Statistics –

  • Taught theory classes, question sessions, and exercise resolution.
  • Developed didactic material and activities supporting the learning process.
  • Covered point and interval estimation, statistical hypothesis testing, chi-square tests, analysis of variance, correlation, and linear regression.

Nodis — StoneCo Ltd group –

Rio de Janeiro, Rio de Janeiro, Brazil · stoneco.com.br

Data Science Intern

  • AI team: built supervised and unsupervised machine learning models, data analysis, and visualization — Python, PostgreSQL, pandas, NumPy, scikit-learn, Matplotlib, Seaborn.
  • Catalog team: maintained the global product catalog and catalog health indicators, scraper mapping, database tables and views — Python, PostgreSQL, MongoDB, Elasticsearch, Kibana, Data Studio.
  • Performance data team: developed hypothesis tests using mathematical and statistical tools — Python, PostgreSQL, pandas, NumPy, scipy.stats, Plotly, Data Studio.

Publications

Peer-reviewed work.

  1. Masid, M. R., Assis, G., Vianna, D., Paes, A., da Silva, A. S.

    Auditing the Evaluators: How Far Can Automatic Evaluation Go in Assessing Portuguese Financial Texts?

    PROPOR 2026, Salvador, Brazil, pp. 222–233.

  2. Marinho, M., Vianna, D., Migliorin, G., Real, L., Ramalhete, M., da Silva, A.

    Grounding the Evaluation of AI-Generated Legal Answers: The HermEval Protocol and Empirical Evidence

    ICAIL 2026, Singapore. in press

  3. Assis, G., Vianna, D., Real, L., Masid, M. R., Nepomuceno, J., Rottschaefer, E., da Silva, A. S., Paes, A.

    Do LLM Judges Agree with Humans? Evaluating Financial Commentaries from Material Facts

    STIL 2026, Cuiabá, Brazil, October 2026. to appear

Education

M.Sc. in Computer Science –present

Institute of Computing, Universidade Federal Fluminense (UFF)

Rated 7/7 by CAPES, the top grade in Brazil's national graduate programme evaluation.

B.E. in Production Engineering –

School of Engineering, Universidade Federal Fluminense (UFF)

Thesis: Streamings in Brazil: characterization and considerations of algorithms.

Exchange Programme –

Kaplan International, Vancouver, Canada

Skills

NLP & generative AI
LLMs, LLM-as-a-judge evaluation, named entity recognition, entity linking, information retrieval, embeddings, BERT-based architectures, text summarization, text similarity
Machine learning
Supervised and unsupervised learning, model evaluation, exploratory data analysis, statistical hypothesis testing
Knowledge & data
Knowledge graphs, semantic platforms and APIs, metadata extraction
Engineering
Python, SQL, Java, PostgreSQL, MongoDB, Elasticsearch, spaCy, Prodigy, scikit-learn, pandas, NumPy, SciPy, Streamlit, Git
Cloud & infrastructure
Google Cloud Platform, Amazon Web Services, Docker
Visualization
Matplotlib, Seaborn, Plotly, Kibana, Data Studio

Software

Writing

Community

Certifications

Languages

Portuguese (native) · English (professional working) · Spanish (professional working)