This repository contains the paper presenting the algorithm Maverick that I developed in my internship as a Machine Learning Engineer at Vela Partners. The model is published on the AI community Hugging Face.
Abstract
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.
Maverick is developed through a transfer learning approach, by fine-tuning a pre-trained BERT model for type 1 and type 2 classification. In this paper we present Maverick, its development, and its performance. 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.