Data Science

SQL Python Tensorflow Spark Git Docker

Description

---

I develop solutions using Machine Learning to generate ingishts through historical data and for application in innovative projects, following the CRISP-DM methodology from extraction and exploratory analysis of data to its deployment

Projects include regression predictions, classification, clustering, time series, natural language processing, anomaly detection, deep learning, computer vision, auto-encoders, Generative AIs (GANs), etc.

Issues such as good programming practices, data cleaning, LGPD, security, Big Data, cloud applications, APIs and deploys are considered throughout the project

Examples

---

• R&D area - Stability prediction of new perovskite compounds
• Financial Area - Forecast and detection of anomalies in financial expenses
• Marketing Area - RFM analysis for customer segmentation
• Risk Area - Credit analysis
• Engineering - Prediction of roughness in aluminum parts after machining
• Teaching - Gerbot: Chatbot that corrects grammar and spelling errors in German
• Medicine - Decision support system for diagnosing peripheral facial paralysis
• Advanced - Generative AI that mimics a Twitter user's writing style.

Data Engineering

Azure Data Factory DataBricks Spark Data Warehouse Data Lake

Description

---

I develop Big Data and data architecture solutions through the Azure platform with design, building, deployment and maintenance of data flows, guaranteeing data reliability

Projects include ingesting data from various sources, such as databases, APIs, sensors and images (combined with machine learning), transforming and storing the data, as well as making it available in order to obtain new insights

Examples

---

• Batch Data - RFM Analysis for Customer Segmentation
• Streaming Data - Coming Soon...

Python

Selenium SpaCy Pandas NumPy Bokeh Others

Description

---

I develop scripts in Python to solve problems with clean, structured and easy-to-interpret code

Projects can include process automation, web scraping, object-oriented programming, parallel programming, Flask APIs, graphical inputs and outputs, data manipulation, advanced calculations, machine learning, cloud deployment, etc.

I also contribute with existing codes, module and API integrations, bug resolution and case testing

Examples

---

• Password Vault
• Personal Wiki
• Weibull Analysis

House-Brackmann

Computer Vision Classification

Links

---

• Presentation - The Developer Conference (TDC)

Description

---

Problem: Peripheral Facial Palsy (PFP) can cause various complications for patients who have it, such as difficulties in feeding or limitations in expressing themselves. In addition, due to the Sars-CoV-2 pandemic, diagnosing PFP becomes even more difficult with the patient's need to travel to a medical clinic.

Solution: The aim of the project was to estimate the House-Brackmann (HB) index of a patient with PFP, for the purpose of supporting medical decision-making. This index indicates the degree of disease severity. The project was guided by a specialist doctor in the field of facial paralysis throughout its development.

Credit Analysis

Classification Machine Learning

Links

---

• Github

Description

---

Problem: Credit analysis is crucial for determining credit card limits, interest rates, payment terms, etc.

Solution: This project aimed to perform credit risk analysis for customers of a store using a cognitive API that communicates with a Sklearn-based machine learning model structured using Pipelines. The API visualization was performed using the Gradio library and all elements were integrated using docker requests.

Results: The project successfully created a scalable, structured, and maintainable credit risk analysis system.

The code was restructured for demonstration purposes and a public dataset was used to preserve confidential information.

Surface Roughness Prediction in Machined Parts

Regression Material Engineering

Links

---

• Poster

Description

---

Problem: The prediction of final material conditions after machining is extremely important due to the slow, expensive, and trial-and-error nature of experimentation.

Solution: The goal of this project is to predict the surface roughness of metallic materials after milling processing by evaluating relevant features and studying various machine learning models, with MLP being selected for implementation.

Results: The project successfully utilized machine learning to predict surface roughness of metallic materials after milling, providing a more efficient and effective approach for predicting final material conditions in the industry.

Nixe Chatbot

Natural Language Processing GPT-3 Audio Processing Chatbot

Links

---

• Github

Description

---

The Problem: Many foreign language students often face challenges when it comes to practicing their writing skills. This is not only due to the difficulty in finding a suitable language partner, but also due to their insecurity when making mistakes while conversing with a real person. As a result, gaps in learning may occur in this context.

The Solution: The solution was to build a foreign language chatbot using OpenAI's GPT-3, Telegram, and Python. The user inputs their speech using Telegram's voice messaging feature and the audio is converted to text. The resulting text is then sent to GPT-3 for processing, which generates a response in the target language. The response is then converted to speech using a text-to-speech library and sent back to the user via Telegram. The conversation history and user information were stored in a SQLite database.

The Results: The resulting chatbot provided a fun and interactive way for the beginner German language learner to practice the language. The use of GPT-3 allowed for natural and diverse conversations, while the integration with Telegram made it convenient and accessible for the user. Storing the conversation history and user information in a database added an extra layer of customization and personalization to the chatbot experience.

Feel free to reach out to me if you would like to learn more about this project or to experience it for yourself.

Product Classification

Natural Language Processing Classification

Links

---

• Open Database: Siscori - Dados Abertos da Receita

Description

---

Problem: The manual classification of imported products in electronic spreadsheets is time-consuming and requires a lot of effort from the operator.

Solution: A cognitive API was developed to automatically classify imported products through natural language processing and Deep Learning. This API uses the open revenue data to access a set of information about Brazilian imported products and classify them according to their HS codes and product descriptions.

Result: For companies that import, monitoring this information and its automatic classification through the API can be essential for effective product management.

Note: Due to privacy issues, additional information cannot be provided.

Thermodynamic Stability of Perovskite Compounds

Materials Engineering Regression Classification

Links

---

• Presentation

Description

---

Problem: Materials Engineering faces challenges in accurately predicting the characteristics of new materials due to limited data.

Solution: I applied several machine learning techniques, to analyze and predict the thermodynamic stability of perovskite materials based on their energy formation.

Results: The method showed successful predictions with high accuracy, and identified extra trees as the best model for both classification and regression. This solution offers a more efficient and effective approach for predicting the characteristics of new materials in the industry.