LeandroGregorini

AI sentiment analysis of sherlock holmes books

Scraped sherlock holmes book texts from Gutenberg Project. After scraping we created a pipeline that analyzes each sentence in each chapter of the books, the model returns a sentiment score and emotions detected in the sentences. For the sentiment analysis the DistilBERT model was used and for the emotion classification the BERT base model was used. After the analysis we visualized the data.

Recommender System

Built a complete beer recommender system based on scraped data from beeradvocate.com. I scraped over 8 million beer reviews and then trained a neural collaborative filtering (deep learning approach) to create a recommender system that can recommend new beers to users based on peer reviews.

Reinforcement Learning

Implemented a Reinforcement Learning project for the popular game Super Mario Bros. The goal was to successfully pass the game using Reinforcement Learning techniques with help of tools available by OpenAI gym. We managed to pass level 1 and 2 by using a double deep q learning approach.

Detection of roundabouts with Deep Learning

Using Deep Learning we implemented several models for the detection of roundabouts on satellite images. We used different computer vision approaches do augement our data and improve model performance. We implemented a multi-input model to further improve predictions.

Eye Tracking Dashboard

Developed a dashboard for visualization of eye tracking data that was captured while volunteers where looking at a metro map. The goal of this project was to visualize collected eye tracking data to understand how metro maps can be improved in terms of better navigation.

Big Data analysis of Bitcoin Blockchain data

We collected every transaction on the Bitcoin Blockchain using a Bitcoin full node. We parsed the Blockchain data and stored the millions of datapoints in Apache Spark. Finally we prepared the data via pyspark and visualized the data.

Unsupervised Machine Learning

Data analysis and different clustering algorithms on popular Palmer Penguins dataset. Clustering of penguins species based on attributes given in the dataset.

Lid driven cavity flow simulation

Flow simulation for the popular lid driven cavity problem in 2D. The flow simulation was realized in python. The finite difference method of Alexandre Joel Chorin (1968) was used to solve the time-dependent Navier-Stokes equation for an incompressible fluid. The simulation was tested on the Ghia et al. benchmark.