Vinicius Brenck

I developed a full-stack web application for Brazil's event hosting and production company. The web app consists of a management software that fully controls the business, its clients, suppliers, and financial structures. It also offers multiple custom reports for the administrator user, including balance sheets, individual profit margin reports for each event, future cash flow analyses, and payables and receivables statements.

The app was developed from scratch to deployment using Django and HTMX. I was in charge of creating the custom financial reports and the banking statements section.

I created an anomaly detection model as the final study for my postgraduate specialization degree in data science and big data.

Using publicly available datasets in the Brazilian government's transparency portal, www.portaltransparencia.gov.br, the model detected around 3,000 anomaly transactions among 4.9 million entries between 2018 and 2020. I used three unsupervised learning anomaly detection algorithms to compose a final voting model: k-NN, isolation forest, and autoencoder. The anomalies detected do not imply fraudulent or irregular transactions but payments substantially different from most government transfers to private entities. The system considers entity and transaction features for the anomaly score.

I built an investment screener app for monitoring real estate funds, known as REITs in the US. The screener uses web scraping to gather information from specialized sites and publicly available data to create and send email alerts of trading opportunities for a small group of investors, according to predefined strategies.

I created an app that runs a scheduled script to monitor public companies' data. It makes HTML emails with trading alerts based on fundamental analysis indicators, including the dividend payout ratio, free cash flow, dividend yield ratio, debt-to-equity ratio, and return on equity. The data is obtained by web scraping publicly available balance-sheet information. After that process, the alerts are submitted to a small group of investors.

I developed a social media marketing app using Selenium to automate an Instagram profile and its posts by managing messages, followers, and likes. The automation provides charts informing the possible origins of new followers and allows programming strategies for obtaining new interactions according to specific business rules. It also includes a plugin to identify if the related parties are a most likely business or personal accounts by checking the profile picture using face detection with a multi-task, cascaded, convolutional neural network (MTCNN).

I obtained 16th place in a Kaggle machine learning competition to develop a model to predict client churn for a telecommunications service. Based on preexisting labeled data, the model uses a supervised learning algorithm of choice to indicate whether clients are prone to churning.