Hi there 👋, welcome to my portfolio. Here you will find links to the Data Science projects I have been working on. The purpose of these projects is to demonstrate my skills in solving business problems using techniques and tools of Data Science.
I am a data scientist experienced in developing business solutions, from the understanding of the business problem to interpreting the model results in terms of business value.
Skills:
Artificial Intelligence: Machine/Deep Learning; Quantum Machine Learning; Large Language Models; Natural Language Processing
Science: Physics: Quantum Physics, Relativistic Physics; Mathematics: Functional Analysis, Differential Equations
Programming Languages: Python; JavaScript; C++
Technologies: Qiskit; Git; Docker; LaTeX
Languages: Portuguese, English, German
The stores of the Rossmann drugstore chain need to be restored and the CEO needs to decide how much is going to be dedicated to the restoration of each one. To support this decision, the Analytics team is asked to present a sales forecast for each store during a period of six weeks, alongside with the total income expected in the chain. This forecast also informs the CEO which store is able to account for its own restoration with the income within this period.
The gross expected income of the majority of stores is in the range between R$5000.00 and R$22000.00. The chain is expected to obtain R$289,822,112.00, with best and worst case scenarios of R$290,808,412.17 and R$288,835,860.27, respectively. These scenarios are predicted using statistical errors (mean absolute percentage error).
A health insurance company intends to offer its customers a new product, a vehicle insurance. In order to achieve this purpose efficiently, it gathered some information about their customers and asked if they would be interested in purchasing a new vehicle insurance. This information was passed on to a Data Science Consulting office.
The office delivered a report informing, among all features gathered, the most relevant ones and the probability of purchase from each customer. Qualitatively, the predicted probability provides a lift gain of 2.5, thus reducing the sales cost to 40%.
Scientific Papers Classification
ArXiv is a public repository of scientific papers where researchers and students from anywhere in the world can find the latest results in many disciplines ranging from Natural Sciences to Mathematics and Computer Science. The papers are categorized by discipline and (possibly multiple) subdiscipline. The platform also displays each paper's abstract without the need to download the file.
The categorization of a new paper is important for authors to make sure their research will reach the intended audience and for the readers to find the most relevant works in their interest area. Presently the categorization process is the authors' sole responsibility. This fact raises some questions:
- Is the category chosen for a research paper the most appropriate?
- Is it possible to make a reasonable prediction of a paper's category given only a summary or it's abstract?
In this project we display how a Large Language Model can be leveraged to help classify a scientific paper. We will employ the transfer learning technique using a pretrained transformer model to predict paper's categories.