Bottomline
March 2022
We all live in a society that produces an overwhelming amount of information daily. Information per se is valuable but it's often very challenging to spotlight the essential part of it - the bottomline, so to say. This mental-filtering process can be very time consuming and also confusing sometimes.
With our technical solution, we provide an automated service that identifies the text's most relevant sentences so as to summarize the text. Additionally, the service provides the general sentiment (positive, neutral or negative) of the text. In other words, the final product will give the user a general idea about the text content as well as its most prominent sentiment.
Tools:
- Python.
- Git, Github, Gitlab and Linux.
- Machine Learning and Deep Learning models for text summarization and sentiment analyses.
- FastAPI.
- Docker.
- Google Cloud Platform and Heroku Cloud.
- MongoDB.
- Streamlit.
Fraud Detection
December 2021
Blocker Fraud Company is a company specialized in the detection of fraud in financial transactions made through mobile devices.
The company is expanding in Brazil and, to find new customers more quickly, it has adopted a very aggressive strategy.
The strategy works as follows:
- The company will receive 25% of each transaction value that was correctly detected as fraud;
- The company will receive 5% of each transaction value that was detected as a fraud despite being legitimate;
- The company will return 100% of each transaction value that was detected as legitimate despite being a fraud.
The final solution includes a Power BI reporting dashboard with answers to business questions as well as a Docker container with API implementation, made with FasAPI and PySpark, and a MongoDB database with APIs requests saved for future analyses. The estimated profit using this solution is BRL 230,133,584.05.
Tools:
- MongoDB.
- Spark.
- Git, Github, Gitlab and Linux.
- Classification machine learning algorithms.
- FastAPI.
- Docker.
- Power BI.
Insiders Project
November 2021
The All in One Place company is a multi-brand outlet company that sells second-line products of several brands at a lower price through e-commerce.
Within just one year of operation, the marketing team realized that some customers buy more expensive products with high frequency and contribute to a significant portion of the company's revenue.
This project aims to determine who are the customers eligible to participate in the Insiders program. Once this list is ready, the Marketing team will carry out a sequence of personalized and exclusive actions to this group of people to increase their sales and purchase frequency.
The final solution answers business questions, validates business hypotheses, creates a Metabase reporting dashboard and implements a solution architecture in the AWS cloud.
Tools:
- SQL, SQLite & MySQL.
- Python.
- Git, Github, Gitlab and Linux.
- Clustering machine learning algorithms.
- Airflow.
- AWS and Streamlit Cloud.
- Metabase.
Sales prediction
October 2021
Rossmann is a company that operates over 3,000 drug stores in 7 European countries. Its products range includes up to 21,700 items and can vary depending on the size of the shop and the location.
Rossmann store managers need daily sales predictions for up to six weeks in advance so as to plan infrastructure investments in their stores (will the next six weeks' sales be high enough to balance infrastructure investment?).
The final solution for this problem is a Telegram bot where the user just needs to type the number of the store and the bot will quickly answer the sales prediction for this given store in the next six weeks.
Besides, if the final user wants more detailed information about this six weeks prediction, he (she) could get further details on a Streamlit data App, with an interactive chart, on sales prediction over these six weeks.
Furthermore, on this data App, the user can also read the entire project overview to understand further how this prediction is made.
Tools:
- Python.
- Git, Github, Gitlab and Linux.
- Forecasting with regression machine learning algorithms.
- Flask API.
- Heroku and Telegram bot.
- Streamlit.
Health Insurance Cross-Sell
September 2021
Insurance All is a health insurance company and its products team is analyzing the possibility of offering a new product, automobile insurance, for its health insurance clients.
Similar to its health insurance, customers of this new insurance plan would have to pay an annual plan to be insured by Insurance All in case of an eventual car accident or damage.
In this project, I developed a Machine Learning algorithm that increases the number of contacted interested customers by 1,316 and 2,259 for 20,000 and 40,000 sales teams contacts so that the estimated revenue increases are respectively U$ 131,600 and U$ 225,900.
Tools:
- SQL, Postgres.
- Python.
- Git, Github, Gitlab and Linux.
- Rank-to-learn machine learning algorithms.
- Flask API.
- Heroku and Google Sheets.
