Developing analysis tools for Esker commits based on GitHub GraphQL API.
Data science and visualization, software development in Python.
Using: Python 3.10 (numpy, pandas, matplotlib, seaborn), conda, GitHub GraphQL API, GitHub Actions, Elasticsearch, Kibana. I deployed a custom Ubuntu server on HyperV with the service.

Founder member of the Kryptosphere branch at INSA Lyon. Treasurer and technical pole manager. I'm in charge of several positions.

Python library development, Jupyter Notebook and Python programming, Web Scrapping, Data Science & Data Classification, Machine Learning, Natural Language Processing, Feature Engineering, Heuristic development.

This internship was composed of 2 distinct projects of approximately 5 weeks each. The first project was to develop a complete Python library to automate the web-scrapping of online newspapers. This task requires solid knowledge in Python and OOP. I implemented and reused results from the following paper: "Détection automatique des liens d’articles dans la une des journaux en ligne" [see in medias]. Since we have to tread a huge number of links, the idea is to use a combination of ML, heuristics and other NLP techniques to try to determine which URLs are interesting and lead to real articles. The combination helps us to increase both precision and recall.

The second part of the project was to work on Data and problem provided by Infolegale. I had to work on outlier detection in a multi-labels noisy imbalanced classification problem. I spent a good amount of time to read scientific papers to familiarize myself with the subject. I then developed my own features, and implemented 4 distinct approaches to solve the problem on a simplification of the initial issue (multi-class noisy classification). I had to discover and use many libraries such as Cleanlab, Pandas, Scikit-learn, Pytorch, Numpy, Pandas, Seaborn. I used many different concepts such as Classifiers, Word vectorizers, TF-IDF, Cross-validation, noise and correlation matrices and many other tools. Finally my best approach worked best than the Cleanlab standard version.

I discovered many things during this internship and made original results on this problem. I want to thank my tutor Előd Egyed-Zsigmond, researcher at LIRIS and his assistant Cédric Bosher for their support.