of
Hi! I'm Renata — a curious mind who turned spreadsheets into dashboards and questions into models. With over 15 years of experience connecting data with real business decisions, I started my journey in the Tax domain, where I created performance dashboards, trained teams, and simplified workflows through automation — even before I knew what ETL was! When I became a mother, I decided to reinvent myself. I moved into tech and fell in love with programming, data storytelling, and Machine Learning. Today, I bring together my business background, analytical mindset, and developer skills to create real impact. Whether building dashboards or predictive models, my goal is the same: make data clear, human, and powerful.
A little about myself.
“I am a tenacious and lifelong learner, fueled by enthusiasm to collaborate with the company and drive toward achieving exceptional results.”
Welcome to my portfolio — a collection of data-driven projects that demonstrate my ability to transform complex problems into actionable insights through Python, Power BI, and Microsoft Fabric.
These projects span Machine Learning modeling, regression analysis, time management optimization, clustering with PySpark, and advanced data storytelling — all built with libraries such as pandas, numpy, matplotlib, seaborn, scikit-learn, and skmultilearn. Each initiative reflects not only technical skill, but also a real-world focus and business context.
I’d like to thank Data Science Academy for their guidance and inspiration throughout my learning journey.
If you’re curious to explore dashboards, notebooks or code, you’ll find the links to my GitHub and project pages below. Enjoy the data-driven dive!
A Data-Driven Journey to Mastering Time Management
Clusterization Analysis in Microsoft Fabric
Real-Time Data Engineering with Microsoft Fabric – BikeTelemetry Project
Linear regression with wheat commodity. - Este projeto está em português.
This project aims to identify the presence of amphibians in the vicinity of two road construction sites in Poland. Our objective is to predict the presence of various types of amphibians near these construction sites and assess the environmental impact on these animals.
.jpg)
The dataset provided by the UC Irvine Machine Learning Repository includes information on buildings, reservoirs, urban or rural areas, reservoir surface characteristics, water reservoir types, the number of reservoirs, minimum distances from buildings and roads, reservoir maintenance status, and shore types.
One of the initial challenges was understanding the different types of amphibians provided in the dataset. As I lack expertise in amphibians, I had to familiarize myself with the fact that the dataset doesn't just include frogs but also various types of frogs, newts, and toads. Additionally, it was important to recognize that each species prefers a specific type of environment.
Another challenge was comprehending the information in the dataset and understanding the meaning of each feature. Upon closer examination, it was surprising to find discrepancies between the expected values and the actual values provided. For instance, for the variable "TR," we expected values ranging from 'a' to 'j' or from 1 to 10. However, the dataset included different numeric categories such as ‘1’, ‘5’, ‘12’, ‘14’, ‘15’, ‘11’, ‘2’, and ‘7’ for the TR feature, with no accompanying information regarding their meanings. These numbers likely represent internal codes used by the authors of the dataset to denote water types, but without a clear key, it was challenging to interpret their significance. This lack of clarity hindered our ability to explore certain features in more detail. Such preliminary analysis is crucial before constructing a model.
In our feature engineering process, we performed numerical and categorical transformations and created new features and classifications to aid in the analysis of amphibian types. Before constructing our model, we visualized relationships and constructed charts to better understand the data. For this project, we utilized the Binary Relevance scheme with a Random Forest model, assuming that all class labels are independent of each other. We employed pandas for data manipulation, matplotlib and seaborn for data visualization, and sklearn for preprocessing the data using MultiLabelBinarizer. Additionally, we built a model using skmultilearn’s Binary Relevance technique with a Random Forest model. Our workflow involved understanding the data, exploring variable categories, transforming variables, and optimizing memory usage. We addressed missing values and unique value considerations, converting categorical types to binary representations.
Analysis and ResultsThe model achieved a Total Hamming Loss of 0.25 and an F1 Score of 0.69. This result is satisfactory, especially when compared to other schemes such as Label Powerset and Classifier Chain, which achieved 29.76% and 27.38% respectively. Our ROC curve indicates good coverage, with values close to 1. Thus, the Random Forest with Binary Relevance Scheme demonstrated superior performance for our classification problem.
This project marked my first experience with multilabel classification. As a result, I now feel more comfortable tackling other multilabel classification problems. I am confident that this project has significantly contributed to my knowledge in this area.
You can see details about this project on my page on the Medium site.Clique aqui
Bike-sharing systems have become an essential part of urban transportation. Accurate predictions of bike rental demand can help optimize fleet management, improve user experience, and ensure a stable supply of rental bikes.
Our dataset comprises hourly records, including weather data (temperature, humidity, etc.) and the total number of bikes rented. Each observation represents a specific hour of the day.
Our goal is to build a predictive model that accurately estimates the bike rental count for each hour based on various factors. This project is not finished yet. What you will see on my GitHub page is a work-in-progress.
To achieve this, I used Linear Regression. The user inputs data about the calendar, weather, and hour, and. The trained model processes this information to estimate the number of bikes available.
During my postgraduate studies in Data Science, I found myself overwhelmed by complex subjects like mathematics and statistics. I realized that my biggest challenge wasn’t just the content—it was the way I was managing my time. I wasn’t studying efficiently, and I needed a smarter approach.
Like many learners, I struggled with distractions and an uneven study schedule. Traditional calendars and to-do lists weren’t cutting it. I wanted more than reminders—I wanted real insights into how I was spending my time and whether I was allocating it wisely across subjects.
By transforming my personal challenge into a professional project, I not only optimized my study routine but also created a powerful portfolio piece. It’s a practical example of how data can improve everyday life—and how technology can support continuous learning.
The TimeManagement Dashboard is a personal initiative focused on time management and study routine tracking, developed using Power BI, Microsoft Fabric, and Microsoft 365.
Because this project contains sensitive data and insights that reflect my personal journey, the repository remains private on GitHub. However, I’m open to sharing access with individuals who are genuinely interested in exploring the structure, visualizations, and solutions applied.
📩 If you’d like to view the project, feel free to reach out to me directly. I’ll be happy to generate a temporary access token with read-only permissions so you can explore the dashboard and related materials.
Access is granted selectively and securely, with the goal of preserving my privacy while encouraging meaningful exchanges with professionals, students, and enthusiasts in the field.
This video series documents the technical journey I undertook to optimize my study schedule through data engineering and visualization tools. Each video highlights a unique stage of my process.
I used Microsoft Graph and Graph Explorer to access calendar data from Outlook. Queries were refined using Postman, ensuring correct time zone settings. The results were saved as a JSON file and uploaded to a Lakehouse in Microsoft Fabric.
I introduced my data workspace and structured raw calendar data using Bronze, Silver, and Gold layers. I developed a function to clean and transform data, applied basic feature engineering, and converted a DataFrame to JSON for use in PySpark.
Defined a schema and created a PySpark table with key event data. Performed transformations and validated the dataset—then saved the output to the Silver layer and prepared it for enhancement in the Gold layer.
Extracted large datasets using a PySpark notebook. Created new columns and performed category-based analysis to uncover insights on time usage, calendar events, and data patterns.
Transitioned from a unified calendar to separate ones focused on studying and learning English. Uploaded JSON files, processed them using PySpark, and refined the setup with a clean, three-step method—later combining the tables in Power Query.
Started with SQL analysis in Microsoft Fabric and moved into Power Query transformations. I selected relevant columns, generated new ones, and validated the transformation logic for calendar event data.
Performed detailed transformations using Dataflow Gen2—changing column types, extracting strings, filtering rows, renaming fields, and building targeted datasets for Power BI dashboards.
Connected Power BI Desktop directly to my Lakehouse in Microsoft Fabric. Imported tables to begin analysis and created visualizations based on structured calendar and time usage data.
First page of my personal time management dashboard — built with Microsoft Fabric and Power BI This project helps me track and optimize my study routines. Still in progress, with Planner integration and more data updates coming soon!
This notebook explores different ways to group data into meaningful clusters, helping to identify patterns and structure within complex information. The goal is to find the most effective clustering model—one that balances accuracy, efficiency, and practical usability.
To achieve this, we tested several widely used approaches, including K-Means, Bisecting K-Means, and Gaussian Mixture models. By comparing these methods, we aimed to understand which one performs best in organizing data meaningfully.
Throughout the analysis, we evaluated key indicators that measure how well the clusters are formed, such as the Silhouette Score and Davies-Bouldin Index. Additionally, we assessed processing times to determine how efficiently PySpark handles clustering tasks.
All of this work was executed within Microsoft Fabric, a unified analytics platform that enabled scalable data processing and simplified orchestration of clustering experiments.
This notebook serves as a comprehensive overview of our findings, offering visual representations, comparisons, and insights into how these clustering techniques behave under different conditions. Ultimately, the goal is to provide a clear understanding of the strengths and trade-offs of each model, supporting informed decisions when applying clustering to future challenges.
In this project, I orchestrated a full-cycle real-time analytics pipeline for urban bike telemetry using Microsoft Fabric and KQL. The initiative involved designing and implementing a robust event house to stream incoming data, refining ingestion strategies, and engineering transformations to ensure data quality and reliability.
Bikes vs BikesCleaned) — with zero data loss post-fix.This hands-on project showcases my proficiency in real-time data flows, temporal diagnostics, and query optimization using KQL — alongside practical problem-solving in stream ingestion and lakehouse architecture.
This dashboard was created using data from the Bicycling EventStream sample available in Microsoft Fabric. Although the dataset is simulated, it mirrors real-time behavior from actual bike-sharing systems—making it ideal for testing ingestion flows, building dashboards, and writing advanced KQL queries.
The dashboard is divided into two pages. In this first section, you'll find a breakdown of data records, including ingestion volume over time, distribution by neighborhood, and detailed record patterns. These visuals demonstrate how real-time data—whether simulated or live—can be transformed into actionable insights for operational analysis and decision-making.
In this second chapter of our Bicycle EventStream Dashboard, we shift our focus from ingestion volume to actionable patterns of behavior. Powered by Microsoft Fabric and real-time streaming data, this section dives into how bike dock availability fluctuates across London — from quiet neighborhoods to mobility hotspots.
Using two dynamic visuals, we reveal the operational pressure points:
These insights go beyond observation. They form the backbone for decisions in redistribution, station expansion, and fleet optimization — making the case for data as a frontline urban planning ally.
Technical proficiency across data science, programming, analytics and enterprise tools.
Xcelis, Remote
RThomaz Tecnologia de Dados MEI, Sao Paulo, SP
Project: Price of Wheat Commodity around the world , Sao Paulo, SP