MeetRox
Data Scientist
-
Rafael Greca Vieira
Machine Learning Engineer and Data Scientist
São Lourenço, Brazil
About
I'm a self-taught and dedicated professional passionate about Artificial Intelligence. I actively seek out new challenges, stay current with the latest advancements, and work on personal projects to deepen my knowledge. My main areas of interest include Natural Language Processing, Deep Learning, Machine Learning, MLOps, and Python. Currently open to remote or on-site opportunities across Europe as a Machine Learning Engineer or Data Scientist, eligible to work in the EU without visa sponsorship.
Interests
Outside of Work
Work Experience
Driva
Data Scientist
-- Reduced false positives by 21% and improved accuracy by 14% in an e-commerce classifier model using a tree-based algorithm and a streamlined data pipeline with PySpark, cutting processing time and memory usage by 33%.
- Created a Python/AWS model with Scikit-learn to match LinkedIn profiles to CNPJs (Brazilian company registry), achieving 83% precision, reducing manual effort, and improving a core product.
- Built a machine learning model utilizing Python, AWS, and Scikit-learn to categorize e-commerce websites into relevant segments, helping a client identify new leads and connect with potential customers.
- Used prompt engineering with the OpenAI API to support data labeling and validation, ensuring high-quality datasets for training ML models.
- Served as tech lead for an intern, providing 1:1 mentorship, teaching core data science and AI concepts, guiding model development, and aligning work with business objectives.
- Developed a data engineering solution to verify a company's location by matching its CNPJ and CNAEs to ANEEL's geographic distributor database using a HEX approach.
CPQD
Machine Learning Researcher
-- Improved a voice activity detection system, increasing accuracy by over 20% and reducing false positives by 12%, thus enhancing the performance of the company’s speech recognition product.
- Developed a supervised sentiment analysis model using a combination of artificial neural networks and a tree-based classifier, enabling actionable insights into user satisfaction with client products.
- Engineered a tree-based text classification model to analyze product reviews, providing insights to guide client business decisions.
- Designed an unsupervised clustering model to monitor internal combustion engine conditions and detect early failures. The results led to a paper submission to an international automotive engineering symposium.
- Implemented a C-based sound event classification system, combining ML and signal processing for near real-time inference on Arduino hardware with high accuracy.
- Built a supervised ML model for classifying dental prosthesis quality in an Industry 4.0 scenario, supporting defect detection and improving batch-level production monitoring.
- Collaborated on a model for sleep stage classification, using heartbeat data from wrist-worn devices to enable non-invasive monitoring.
- Authored technical documentation, patent applications, software registrations, and research papers related to the developed models.
Zenvia
Data Scientist
-- Built new machine learning and deep learning models for NLP tasks, such as sentiment analysis, clustering, and topic classification, enabling performance insights for client chatbots.
- Improved data processing speed and memory efficiency by over 4 times by redesigning the data acquisition and preprocessing pipeline using multiprocessing and simplified functions.
- Improved customer engagement by more than 15% by improving the company’s communication channel recommendation algorithm through a statistical, rule-based approach.
- Delivered data visualizations and analytics reports that supported the curatorial team in data-driven business decision-making.
Education
University of São Paulo
Federal University of Itajubá
Selected Projects
Multiagent Architecture for YouTube Short CreationNew
A multi-agent pipeline that automates the creation of YouTube Shorts from a single topic: it generates video ideas, writes and validates scripts, produces voiceover with ElevenLabs, creates thumbnails and scene images with OpenAI, assembles the video with MoviePy, and generates SEO metadata.
End-to-end MLOps Project
The purpose of this project's design, development, and structure is to create an end-to-end Machine Learning Operations (MLOps) lifecycle to classify an individual's level of obesity based on their physical characteristics and eating habits.
ScratchML
A Python library called ScratchML was created to build the most fundamental Machine Learning models from scratch (using only Numpy), emphasizing producing user-friendly, straightforward, and easy-to-use implementations for novices and enthusiasts.
Speech Emotion Recognition using Deep Learning and Discrete Wavelet Transform
A deep learning solution using Convolutional Neural Networks and Wavelet Transform to tackle the Speech Emotion Recognition task. Repository dedicated to the developed solution for the end-of-course work of the University of São Paulo's Master in Business Administration (MBA) in Artificial Intelligence and Big Data program.
Voxseg PyTorch
The non-official Voice Activity Detection (VAD) Voxseg model implementation in PyTorch. Voxseg is a Python library for voice activity detection (VAD) for speech/non-speech segmentation.
Classification of Hate Speech Tweets using Deep Learning Models (Portuguese Only)
Comparison of deep learning models' (Convolutional Neural Networks and LSTM) performance in classifying English tweets containing hate speech. Repository dedicated to the developed solution for the end-of-course work of the Federal University of Itajubá's Bachelor in Computer Science.
Publications
A Speech Emotion Recognition Approach Using Discrete Wavelet Transform and Deep Learning Techniques in a Brazilian Portuguese Corpus
Environmental Monitoring with Low-Processing Embedded AI through Sound Event Classification
Unsupervised Clustering for Internal Combustion Engines Health Monitoring