Startup Success Forecasting Framework (SSFF)

The Startup Success Forecasting Framework (SSFF) is a pioneering approach designed to automate the evaluation of startup success potential. Leveraging a blend of traditional machine learning models, Large Language Models (LLMs), and real-time market data analysis, SSFF aims to transform the landscape of venture capital investment by providing deep, actionable insights into the viability of early-stage startups.

Link to Paper: https://arxiv.org/abs/2405.19456

Project Structure

project_root/
│
├── .streamlit/              # Configuration for Streamlit app
├── agents/                  # Core agent logic (founder, market, product, etc.)
├── algorithms/              # ML algorithms, embeddings, similarity calculations
├── data/                    # Raw and processed datasets (e.g., Merged_Successful_V2.csv)
│   ├── Successful/
│   └── Unsuccessful/
├── EDA/                     # Exploratory Data Analysis scripts and notebooks
├── experiments/             # Experiment configurations, logs, and raw results
│   ├── charts/
│   └── results/
├── models/                  # Trained model artifacts (e.g., .keras, .joblib)
├── plots/                   # Generated plots and visualizations
├── rebuttal_supplements/    # Supplementary materials, including statistical summaries
│   └── M1_Statistical_Summary/ # Detailed descriptive statistics and CSV outputs
├── utils/                   # Utility scripts, configuration, API wrappers
│
├── app.py                   # Main Streamlit application for web interface
├── main.py                  # Alternative main execution script (if applicable)
├── overallPipeline.py       # Core pipeline orchestration script
├── baseline_framework.py    # Scripts for baseline model implementations
├── ssff_framework.py        # Core SSFF framework logic
│
├── requirements.txt         # Project dependencies
├── .env                     # Environment variables (API keys - create from .env.example if provided)
├── README.md                # This file
└── SETUP_GUIDE.md           # Detailed setup instructions

Environment Setup

To set up the environment for this project, follow these steps:

1. Ensure you have Python 3.7+ installed on your system.

2. Clone the repository:

   git clone https://github.com/your-username/Startup-Success-Forecasting-Framework.git
   cd Startup-Success-Forecasting-Framework
   

3. Create a virtual environment:

   python -m venv myenv
   

4. Activate the virtual environment:

   • On Windows:

     myenv\Scripts\activate
     

   • On macOS and Linux:

     source myenv/bin/activate
     

5. Install the required packages:

   pip install -r requirements.txt
   

6. Create a .env file in the project root (you can copy env.example if it exists) and add your API keys:

   OPENAI_API_KEY=your_openai_api_key_here
   SERPAPI_API_KEY=your_serpapi_api_key_here
   

7. To deactivate the virtual environment when you're done:

   deactivate
   

What This Repo Includes

• Data Exploration Files: Scripts and notebooks for exploring and understanding the dataset used in model training (see EDA/ and rebuttal_supplements/M1_Statistical_Summary/).
• Model Training Files: Code for training the SSFF's underlying machine learning models, including random forest classifiers and neural networks (see algorithms/ and specific model scripts).
• Pipeline Building Files: Implementation of the SSFF pipeline, integrating LLM analysis, feature extraction, and prediction models for comprehensive startup evaluation (see overallPipeline.py, ssff_framework.py, and agents/).
• Experiment Results: Outputs from various experimental runs can be found in experiments/results/ and summarized in rebuttal_supplements/.

An image depicting the SSFF architecture. (Note: Original image URL may contain user-specific identifiers).

Updates

A web interface has been developed! Very easy to interact with:

(Note: Original image URL may contain user-specific identifiers).

To-Do List

• Launch a Demo Interface
• Upgrade to Version 2

How to Run

To execute the main pipeline and analyze a startup, you can run the Streamlit web application:

streamlit run app.py

Alternatively, use the main pipeline script (consult main.py or overallPipeline.py for direct script execution details):

python overallPipeline.py # Example, verify correct script and arguments

This framework supports two modes of operation:

• Simple Mode: Provides a quick evaluation based on predefined criteria.
• Advanced Mode: Offers an in-depth analysis incorporating external market data, founder-level segmentation, and custom LLM prompts for nuanced insights.

Reference Paper

For a detailed understanding of the methodologies and technologies underpinning SSFF, refer to our accompanying paper titled "An Automated Startup Evaluation Pipeline: Startup Success Forecasting Framework (SSFF)". This paper discusses the challenges of early-stage startup evaluation and presents our novel framework as a solution that combines machine learning, natural language processing, and real-time data analysis.

Key Highlights from the Paper:

• Startup Evaluation Pipeline: Exploration of traditional and emerging approaches to startup evaluation, emphasizing the integration of qualitative assessments.
• LLM Agent and Prompting Techniques: Insight into how Large Language Models and sophisticated prompting techniques can enhance the evaluation process.
• Founder Level Segmentation: A novel approach to categorizing founders based on their experience and potential impact on startup success.
• External Knowledge Block: Utilization of real-time market data to enrich the analysis and provide a current view of market conditions and trends.

Conclusion

The SSFF represents a significant step forward in automating and enhancing the accuracy of startup success evaluations. By combining cutting-edge AI technologies with a deep understanding of the startup ecosystem, SSFF empowers investors, entrepreneurs, and researchers with a powerful tool for making informed decisions.

Acknowledgments

This work was supported by contributions from project members and a partnering firm. Their invaluable insights and expertise have been instrumental in the development of the SSFF.