Raquel Colares

Intro

AI Specialist and Data Science professional with diverse experience in developing solutions across key areas of artificial intelligence, including supervised, unsupervised, and generative techniques. Experienced in designing and implementing intelligent systems that combine data processing, modeling, and automation to solve real-world problems. Skilled in analytical, critical, and scientific thinking, I am well-equipped to face complex data-driven challenges.
Check out my work.

Skills

• Programming & Tools: Python, SQL, Git, GitHub, OOP
• Libraries & Frameworks: Pandas, NumPy, Scikit-learn, PyTorch
• Machine Learning: Supervised and unsupervised learning, Feature engineering, Model evaluation
• Deep Learning: Neural networks, Convolutional Neural Networks (CNNs), Transformers, LLMs, Multimodal models, Autoencoders
• Generative Models: Variational Autoencoders (VAEs), GANs, Diffusion Models, Autoregressive Models
• Natural Language Processing: NLP pipelines, LLM fine-tuning, Semantic search
• Data Visualization: Power BI, Streamlit, Matplotlib, Seaborn, Plotly
• Other Strengths: Problem-solver, Solid mathematical background, Attentive to detail

Work

Speech-to-Text Interface for Physiotherapists

ROMTech is a multimodal AI interface that I developed to support physiotherapists by transcribing and summarizing clinical consultations. It integrates three core models: a speech recognition model (using Whisper) for converting audio to text, a deep autoencoder trained on sentence embeddings for unsupervised summarization, and a BERT-based model used as a complementary approach. The system processes multilingual audio inputs and provides both full transcripts and concise summaries. All functionalities are delivered through an interactive Streamlit interface that I built to help reduce documentation time and allow physiotherapists to focus more on patient care.

Cancer Classification

This project explores the use of artificial intelligence to support early cancer risk classification through two predictive models: MLP Classifier and XGBoost. By implementing a full pipeline, an API using FastAPI, and an interactive Streamlit interface, users can input personal health data and receive real-time predictions indicating their likelihood of developing cancer. The interface offers a choice between the two models and displays results visually, along with messages emphasizing the importance of early detection and cancer prevention. Both models achieved high accuracy and generalized well to unseen data. While not intended to replace medical diagnosis, the project aims to demonstrate how AI tools can increase awareness and assist in identifying cancer risk early, when treatment outcomes are most favorable.

Classification of Cardiac Arrhythmias using Machine Learning Models: A Data Science Study

This project focuses on the early detection of cardiac arrhythmias, one of the leading causes of mortality worldwide. Using publicly available Holter monitor datasets, the first from the St. Petersburg Institute of Cardiological Technics (INCART) and the second from the Massachusetts Institute of Technology (MIT), a total of ten machine learning models were trained and compared to classify arrhythmic patterns. These included Naive Bayes, k-NN, SVM, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, Bagging, XGBoost, and LightGBM. After evaluating performance using metrics such as accuracy, precision, recall, F1-score, and AUC-ROC, the XGBoost Classifier stood out with the best performance (AUC = 99.91%). Model interpretability was supported by SHAP, which identified the RR interval as the most relevant feature, aligned with clinical knowledge, as this interval is key in assessing cardiac risk. Heart rate variability reflects beat-to-beat changes in RR intervals, which are influenced by the autonomic nervous system, and imbalances in this system are known to be associated with a higher risk of cardiac mortality.
This project was developed as part of my postgraduate studies in Brazil, and while the full document is written in Portuguese, I would be glad to discuss the methodology, results, or potential collaborations, in English or French as well.

About

I'm passionate about using machine learning and data to make things smarter, faster, and more meaningful. I love turning complex problems into clear, data-driven solutions. Most of my projects focus on real-world impact, especially in areas like healthcare and science, where AI can really make a difference. I enjoy working in collaborative environments where I can explore new ideas, solve complex challenges, and keep learning along the way.
Curious by nature and driven by purpose, I bring both creativity and structure to everything I do.

Contact

• LinkedIn
• GitHub
• Email

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';