Felix Ringe - Law and Machine Learning

Determining the minimum base capital present at Annual Shareholders' Meetings (ASMs) based on ASM reports with GPT-4 (December 2023)

This project determines the minimum base capital present in Annual Shareholder Meetings (ASMs) of German companies based on unstructured shareholder reports data, achieving 98.4% accuracy. The test set, representing a comprehensive sample, includes one report from each of the 178 companies assessed, drawn from a pool of approximately 3700 reports.

lectures.berlin (February 2023)

With the aim of bringing the three Berlin universities and Charité even closer together, we have put together a platform that allows students to search for classes from all universities at the same time. Note: This is work in progress as we are still procuring some lecture data.

ECOmpute (December 2022)

A green energy scheduler for compute-intensive tasks that reduces your carbon footprint. The app allows users to schedule compute-intensive, and therefore energy-intensive, tasks in the cloud. By predicting times at which most green energy is available, it reduces carbon emissions particularly by leveraging times at which green energy is generated but not used. Our project is built around the concept of maximizing "green time" - the time in which the percentage of renewable energy used for the compute task is above a high threshold, ideally using only sustainable means of generation.

This project was selected as one of nine finalist projects at LauzHack 2022.

Accenture Challenge: Supply Chain Resilience (November 2022)

Based on a dataset from a fashion retailer in Europe that included geographical and order data, our challenge was to build a classification model that predicts the probability of an order being delayed. We engineered 32 features and created a model ensemble based on Random Forest, Nearest Neighbour, XGBoost, AdaBoost and MLP. The model achieved an accuracy of ~82%.

We built a Streamlit app around it that allows the retailer to provide data on a specific order and have the model predict the likelihood of it being delayed, empowering them to assess the reliability of specific routes and e.g. renegotiating contracts with those third-party logistics partners that have proven to have an increasing effect on an order being delayed.

LSC Sync (July 2022)

An application that fully automates administrative tasks for my job at Laws of Social Cohesion. It maintains the events page of the project.

HackUPC McKinsey Challenge: Time series forecasting in grocery e-commerce (May 2022)

A time series forecasting project built at HackUPC 2022, a 36-hour hackathon in Barcelona. We should forecast how sales would develop across a two-week timespan.