Beatriz Maia
Available for collaboration

Beatriz Maia

Machine Learning Engineer

LinkedIn GitHubiD ORCID

I'm a Machine Learning Engineer focused on the end-to-end training and deployment of AI models. I work across the full spectrum: from building custom deep learning and computer vision architectures to fine-tuning and aligning large language models. My emphasis is on models that are accurate, cost-effective, and secure, using techniques like quantization and distributed training to optimize performance at scale.

Beyond custom modeling, I design production-grade systems that integrate models from OpenAI, Google, and Anthropic into scalable backend infrastructures. My approach bridges research and engineering: I bring insights from academic work to solve real-world problems through thoughtful architecture and rigorous evaluation.

Based in Vitória, Brazil with multicultural experience having lived in both Brazil and the United States. Bilingual in Portuguese and English.

Experience

Professional journey

  1. Apr 2025 – PresentBrazil

    Machine Learning Engineer

    Cadastra (Contracted to Grupo Boticário)

    • Maintains and updates a GenAI Gateway system connecting multiple third-party LLM providers like OpenAI, Google, and Anthropic via Azure and VertexAI.
    • Engineered a feature for the company's GenAI SDK using LiteLLM to allow for model-switching and unified classes across frameworks like Agno, CrewAI and Google ADK.
    • Developed APIs for image/audio transcription and RAG-as-a-service using FastAPI, LangChain, and LLMs.
    • Built an autonomous system on GCP that reduced a manual 1-hour data update process to just 5 minutes.
    • Used various RAG methods, like LLM Embeddings with third-party models, and maintenance of a Cache-Augmented Generation (CAG) implementation.
    • Performed database mapping and complex queries using PostgreSQL and SQLAlchemy.
    • Implemented Unit Testing and automated evaluation pipelines using LLM-as-a-judge to ensure code and prompt quality.
    • Used Kanban methods and metrics for organizing tasks and measuring the team's performance.
  2. Oct 2024 – Nov 2024Brazil

    Machine Learning Engineer

    Docket

    • Fine-tuned multimodal models for Named Entity Recognition (NER) on legal deeds to assist in property ownership data extraction.
    • Used Google Cloud Vision API to extract text from high-variance notary documents.
  3. Nov 2023 – Oct 2024Brazil

    Machine Learning Engineer

    Weni

    • Fine-tuned models up to 70B parameters using PyTorch, DeepSpeed, Hugging Face Accelerate, and PyTorch FSDP.
    • Aligned LLMs using SFT and RLHF/DPO for Zero-shot classification and Q&A tasks.
    • Applied AWQ Quantization to 70B models to make production deployment more efficient.
    • Compared RAG pipelines using LangChain, Hybrid Search and Cohere Reranking logic with Amazon Bedrock.
    • Developed semi-autonomous training pipelines to move experiments from Google Colab to RunPod.io, enabling multi-GPU scaling and increasing the team's MLOps maturity.
  4. Aug 2020 – Oct 2024Brazil

    Researcher

    Bio-inspired Computer and Engineering Lab (LABCIN)

    • Lead Author of research on Transformers and Few-Shot Learning applied to digital pathology and clinical skin images, published in the IEEE Journal: Expert Systems with Applications (2023).
    • Investigated Variational Autoencoders (VAEs) and Few-Shot Learning for medical feature extraction.
    • Peer-reviewed for the IEEE Journal of Biomedical and Health Informatics (ORCID: 0000-0003-1057-2785).
    • Developed model training pipelines using PyTorch, MLFlow and DVC
  5. Jun 2022 – Oct 2023Brazil

    Machine Learning Engineer

    Eldorado Research Institute

    • Participated in research groups for MLOps and Explainable AI (XAI).
    • Developed a hybrid OCR pipeline using OpenCV and Tesseract, and used YOLO for element detection.
    • Developed NLP and Named Entity Recognition (NER) pipelines using Hugging Face to process and extract data from unstructured text.
    • Curated Python-based Docker images to ensure they were vulnerability-free or low risk.
  6. Jun 2021 – Jun 2022Brazil

    Data Scientist (Intern → Full-time)

    VERT Transformação em TI

    • Led client meetings and developed risk analysis frameworks for government bidding processes (State of Pará) and DETRAN-ES financial projects throughout both internship and full-time roles.
    • Architected ETL processes in SAS (9.4/Guide) and managed database queries and mapping with PostgreSQL and SQLAlchemy.
    • Developed web scrapers using BeautifulSoup and Selenium for automated public data collection.

Education

Academic background

  1. Apr 2019 – Expected Mar 2026Vitória, ES, Brazil

    Bachelor's Degree in Computer Science

    Federal University of Espírito Santo (UFES)

    Coursework completed in 2024. Emphasis on Machine Learning and Software Engineering.

Volunteering

Community & service

  1. Aug 2019 – Nov 2022Vitória, ES, Brazil

    Teacher

    Introcomp — Introdução à Computação (Introduction to Computing)

    • Taught Python programming to public high school students and created Jupyter notebooks for use as study materials and teaching guides.
    • Lead the data collection and class streaming committees.
    • Co-authored an article on programming teaching challenges for public school systems for the 2021 Brazilian Symposium on Computing Education.
  2. Sep 2021 – Dec 2021Vitória, ES, Brazil

    Information Technology Support Technician

    Dermatological Assistance Program (PAD-UFES)

    • Provided IT support for a rural healthcare program, assisting with patient registration systems and data collection for skin cancer datasets.
  3. Apr 2019 – Mar 2020Vitória, ES, Brazil

    Laboratory Assistant

    Laboratório de Administração de Redes (Network Administration Laboratory) — UFES

    • Assisted professors with exam applications and managed network and user account issues within the Computer Science labs.