APPLICATION OF STARTUP SUCCESS PREDICTION MODELS AND BUSINESS DOCUMENT EXTRACTION USING LARGE LANGUAGE MODELS TO ENHANCE DUE DILIGENCE EFFICIENCY (CASE STUDY: LIVING LAB VENTURES)

• Main Author: Christian Samudra, Vito
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85142
• Institution: Institut Teknologi Bandung

Startups face extreme uncertainty and high failure rates, making the identification of potential startups a challenge for investors. This research leverages Large Language Model (LLM) and Machine Learning (ML) technologies developed using the Team Data Science Process (TDSP) methodology. The main steps in system development include processing and integrating startup data, developing a Machine Learning (ML) model for startup success classification, and integrating the OpenAI API with the GPT-4 model and Google Search API for business, financial, competitor, and market trend analysis. The developed system's dashboard includes key features such as pitch deck analysis, financial analysis, market trends, competitor analysis, founding team analysis, and startup success prediction. The startup success prediction feature was developed using the XGBoost model, which has shown the best and most consistent evaluation results with cross-validation. The model is then saved in a pickle file and deployed using Flask to interact with the system. Customer acceptance testing results showed an acceptance rate of 4.50 out of 5.00, filled out by eight experienced professionals as startup investors, reflecting a high level of satisfaction with the developed system.