Duke Capital Partners (DCP) is initiating a project to develop an internal sourcing tool aimed at identifying Duke University affiliated startup companies seeking funding, which leverages the power of the Duke network to support the university’s entrepreneurial ecosystem.

Implement advanced filtering system and scoring system which could be directly used upon the raw dataset, enable users (DCP members) to search for Duke-related companies they are interested in (including startups) based on specific criteria such as Duke afiliation, industry, company size, company description etc.

• Filtering system: The goal of this system is to filter out irrelevant features from our raw datasets, and only keep the features that DCP members care about.
• Scoring system: The goal of this system is to score each company on the list based on its features, in order for DCP members to find the companies they are interested in.

• LinkedIn Sales Navigator (used for all model training)
• Techstars Jan 2024 Batch data for description classifier fine-tuning

• Algorithmic scoring model with 5 weighted criteria based on LinkedIn company factors
• Manually evaluated by DCP team for sorted results aligning with their target companies
• Mostly used for filtering out unwanted metrics from the classification metric

• Fine tuning Bert to do a multiclass classification to classify the company description (one of the features of our data) based on part of the collected data with manually assigned labels. (0 = not relevant at all, 1 = startup companies with Duke connections, 2 = worth our time! )
• Preview of training dataset we used:

• Final model takes in augmented LinkedIn data and description classifications for softmax ranking of company relevance

• Company Description Classifier: 0.602 F1-score on validation set
• Neural Network Classifier: 0.84 Recall on positive class predictions
• Streamlit local-hosted UI for simple results pooling

To run this project, follow the steps below:

1. Create a Python Environment

   • If you haven't already, create a virtual environment using your preferred method. For example, you can use virtualenv:

     virtualenv venv

   • Activate the virtual environment:

     source venv/bin/activate

2. Install Dependencies

   • Install the required Python packages by running:

     pip install -r requirements.txt

3. Open Terminal

4. Navigate to the root directory of the project

5. Run the following command:

   streamlit run app.py

6. The UI will open automatically in your default web browser. If it doesn't, navigate to localhost:8501 manually.

   

7. This is how the UI will look like:

   

   • Button to upload CSV data
   • Button to upload the model
   • Button to predict using the data (will appear after the model is uploaded)

8. Upload the data using the first button:

   • Select the CSV data file
   • Click on 'Open'

   

9. Select the model in the same way

   

10. Wait for a while until the documents are processed. There will be a running indicator on the top left.

    

11. After the processing is done, the 'Predict' Button will appear. Click on it and wait.

    

12. Prediction is complete, and you can see the results in the new table

    

13. Click on the LinkedIn Profile, which will redirect you to the link

Feel free to reach out if you encounter any issues!