Cubatorin - The Chatbot Incubator for Investors and Startups
Laily Ajellu, BSc, Jeff Edmonds, PhD, (Supervisor)
School of Engineering,York University
Introduction
Cubatorin is a business incubator for startups. It offers detailed instructions and statistical research to entrepreneurs, based on their industry and their startup idea. The platform also facilitates the writing and sharing of business proposals between entrepreneurs and interested investors. It includes a chat interface for investors to confirm or deny proposals
Entrepreneur’s View
Figure 1. Building their business proposal (AI generated Instructions)
Investor's View
Figure 2. Checking a startup’s business proposal
Results
The key metric that we have monitored is the quality of the detailed instructions and statistical research.
We have evaluated this using multiple methods:
- Investor feedback (Thumbs up/Thumbs down)
- Model metrics (Accuracy and Loss of the AI model in the Training and Validation phases)
The accuracy of the AI has generally improved with more data samples, however, quality matters. Anytime the AI's answers start generalizing again, the most recent data samples are discarded to retain specificity.
Effect of Article Upload on ChatGPT Accuracy
Chart 1. Comparing the base model's accuracy (51%) with Cubatorin’s model accuracy (74%), we can see a long-term improvement.
Number of Positive Ratings vs. Number of Tokens
Chart 2. Investors have higher ratings in the second half of the AI’s training phase.
- 4 positive ratings/105 tokens in the first half
- 5.33 positive ratings/105 tokens in the second half
Problem Being Solved
ChatGPT's responses are often too generalized. Cubatorin provides more specific answers tailored to the field of startup investing. There are many repetitive (yet industry-specific) questions and processes that investors and entrepreneurs need to discuss before closing a deal. Cubatorin is the liaison between the two worlds.
Figure 3. URL/File upload used for training AI
Methods and Materials
Cubatorin makes its answers more specific by fine-tuning an OpenAI chatbot.
Data Sources: The BEST team at York University continuously curates and uploads relevant articles for training the AI. As new startups join the incubator and new, practical methodologies and articles become available, Cubatorin’s instructions are regenerated.
Methodology: The curated data is given priority over the data originally sourced for the base model of ChatGPT (formed from general information on the web).
Discussion
Cubatorin has a direct impact on the York University community as the software will be used at the Bergeron Entrepreneurs in Science and Technology program to help students develop their startups.
Beyond York University, we plan to share it with other incubators and investors to speed up their process of selecting startups to fund.
The insights provided by Cubatorin's instructions to the entrepreneur drastically reduce the amount of time investors must spend mentoring their chosen startups (by about 40% per workday from 2 hours to 1.2 hours). In addition, the research conducted by the AI is based on the startup's industry and speeds up the process of product development.
Figure 4. Feedback given for a Test of the newly trained AI
Conclusion
With Cubatorin, investors spend less time researching and guiding entrepreneurs in their niche industry. In addition, entrepreneurs have more support while building their product.
These results allow investors to allocate more time to reviewing and providing feedback on business proposals. It also provides entrepreneurs with more robust business planning.
Contact Cubatorin
Laily Ajellu: lajellu@my.yorku.ca
York University: Call +1 (647)622-4273