ML Expert/Python Developer

• Sep 2020 - Present

Data Scientist consultant

• Dec 2019 - Jul 2020

Head of one of the 3 biggest projects in the Lab: Java ML Plugins - Supervised and assisted a group of 3-5 people in order to implement the backend part of an AutoML framework on Java. Tools: JIRA, GitLab, Microsoft Teams. - Adapted plugins of machine learning algorithms (PCA, GradientBoosting, RandomForest, etc.) from pre-existent libraries (Smile, Weka, DeepLearning4J) to the Lab’s framework data structure and applied them to solve machine learning problems. - Developed from scratch a Java model of the gradient Boosting algorithm LightGBM. Tools: C++, Python, SKLearn, Java. Side projects: - Experienced with all the preprocessing and classification algorithms in the Splunk platform to solve ML problems (1 month) - Developed on Java a framework able to extract streaming data from an application, process and clean them through an integration between Apache Kafka and Spark Streaming and storing the final information in Cassandra database. (5 weeks) - Studied state-of-the art of Meta-Learning and Auto Machine Learning (2 weeks) - Supported a mission on NLP to scrape information from websites using python "scraper" library Show less

Research Assistant in Deep Learning

• Jan 2019 - Aug 2019

- Worked on Python on my thesis project on Sarcasm Detection on Social Media - Used Twitter API and Python’s library “tweepy” to create a dataset (more than 8 GB), running “allennlp” libraries and Pytorch on AWS to create a model for sarcasm detection - Exploited Deep Learning techniques combining NN models such as LSTM, BiLSTM, BiLSTM with Attention, CNN and Word Embeddings (FastText, GloVe, ELMo and their concatenation) to detect sarcasm in social media - Implemented a MultiTasking Framework to improve the performances of sarcasm detection task through a sentiment detection task (experienced with “pycorenlp” library for Sentiment Analysis) - Reached state-of-the-art results on a prelabelled dataset - Obtained coherent predictions with real word data Show less