Traditional valuation methods often fail to capture the complexity of modern startups which are driven by a variety of dynamic factors. In many cases, traditional approaches to predict startup valuation rely on structured data such as the age of the startup, number of founders, and sector. However, these approaches ignore the weight of qualitative information in unstructured text such as the descriptions of innovation and business models. In this paper, we propose a proof of concept of using a machine learning model that combines multiple data modalities to predict a startup valuation. Considering numerical features, categorical features, and text embeddings derived from unstructured text data using a Large Language Multimodal, BERT (Bidirectional Encoder Representations from Transformers), all features will be integrated into one model classifier. This approach aims to show how textual data can be leveraged with structured inputs to improve the precision and relevance of valuation predictions. This approach has the potential to serve as a foundation for the development of decision-support tools that enable investors to make more informed and data-driven financial decisions.