Gustavo Muller

I developed this algorithm from scratch. There were no AI libraries available at that time, so everything has to be implemented on your own.

The software used machine vision algorithms like image opening, closing, filling, and Sobel filtering to find possible candidates for plate regions. Candidates were classified by aspect ratios and the number of elements.

The neural network was an RNN implemented in C and converted the image characters into text. The algorithm achieved up to 90% effectiveness in recognizing car plates for speed enforcement.

Usage Based Insurance (UBI) was the client and a leader in their field in Brazil, and I led a team of developers and engineers in the development of their platform.

The platform collected data from over 300,000 tracked vehicles and sent the data to AWS servers that processed the data using big data technologies. We had over 1TB of data to be processed into individual driver behavior models. The result was a driving score that measured how prone the car was to get into an accident. The client then sold this information to insurance companies.

The technology involved many clever AI, including neural networks and machine learning algorithms, to achieve this. The end-user was able to consume this data using mobile applications (iOS/Android).

I developed an automatic trading platform to dispatch trading signals when they occur. The idea was to use known trading setups and let the software monitor multiple tickers and timeframes. When the trading condition is met, we then dispatch a trading signal to a website.

I used MetaTrader to implement the logic to detect the trading signals and SQLite to store them locally and a Python application to replicate data to the cloud. Once in the cloud, the data is consumed by a Vue.js-based website using a REST API built with Django.

The app monitors over 100 tickers in five different timeframes each, and we implemented over ten trading setups.