Maverick <br>

Developed during my internship at Vela Partners as a Machine Learning Engineer. <br> The paper presenting Maverick can be found on my GitHub. <br> Maverick consists of two sub-models published here on Hugging Face : MAV-Moneyball & MAV-Midas

Abstract <br> Maverick (MAV) is an AI-enabled algorithm to guide Venture Capital investment by leveraging BERT - the state-of-the-art deep learning model for NLP. Its ultimate goal is to predict the success of early-stage start-ups.

In Venture Capital (VC) there are two types of successful start-ups: those that replace existing incumbents (type 1), and those that create new markets (type 2). In order to predict the success of a start-up with respect to both types, Maverick consists of two models:

• MAV-Moneyball: predicts success of early stage start-ups of type 1.
• MAV-Midas: predicts whether a start-up fits current investment trends made by the most successful brand and long-tail investors, thereby taking into account new emerging markets that do not necessarily already have established successful start-ups leading them - ie. start-ups of type 2.<br><br>

Maverick is developed through a transfer learning approach, by fine-tuning a pre-trained BERT model for type 1 and type 2 classification. Notably, both MAV-Moneyball and MAV-Midas achieve a true positive ratio greater than 70%, which in the context of VC investment is one of the most important evaluation criteria - it is the percentage of successful companies predicted to be successful by Maverick.