Luan Brasil

Development and maintenance of taxes forecasting using Time Series models (e.g. Exponential Smoothing, Naive Forecasting), including new experiments and results analyzis, at a cross-functional team for the Tax Intelligence product. Responsible for delivering taxes forecasting analysis and communicating with stakeholders and managers.

Created an AI-powered Chatbot for data queries via WhatsApp, designing and developing the data architecture in Google Cloud Platform, using BigQuery for Data Warehousing, Cloud Run to deploy the bot API and Twilio to handle messages in WhatsApp. Used OpenAI's GPT models to identify questions from users and convert them into SQL queries so the LLM model would return the results from the query to the user as a message with the requested data. Also, I leveraged a RAG mechanism using Pinecone for vector storage and search to enhance the bot answers and optimize prompts. Built for directors and managers in the administrative, finance and marketing areas from shopping centers and media platforms. Developed several media data analyses, discovering the best advertisement content to increase revenue and business metrics. This is done using Python, AWS (get data from Athena) and GCP (get analytical data from BigQuery), implementing data pipelines to get the data from the data sources and apply data processing techniques to refine it and get valuable insights. These analyses helped optimize media investments for a news and media company. Developed a delivery time prediction product for the logistics teams of many different malls in Brazil, based the roadmap of this product development on CRISP-DM, finishing with a Decision Tree-based regression model to help logistics areas organize their delivery queue and decrease product delays. Created dashboards for various stakeholders about shopping, e-commerce, and media performances using Metabase and Power BI, getting the data from a Data Lake and Google Analytics for a News platform in Brazil.

Built data pipelines with Databricks and BigQuery and ETL process using Python and Spark at a waiting time forecasting for patients in the ER product. Also, applied Machine Learning techniques that decreased the average model error by 30%, using better data cleaning and preprocessing techniques such as outlier detection and categorical feature encoding. Allowing hospitals in Brazil to have a better user experience for their patients in the ER, increasing the total hospital’s NPS (Net Promoter Score).