OskiGPT

Revolutionizing entrepreneurship at UC Berkeley, our AI chatbot provides a personalized gateway to hundreds of resources, enabling innovators to focus on big ideas, not research

Comment

Inspiration

  • As aspiring entrepreneurs at UC Berkeley, we realized the treasure trove of startup resources available to students were scattered and difficult to navigate. The process was time-consuming and often overwhelming - a problem we knew needed to be solved.

Information spread all over numerous Berkeley websites Berkeley entrepreneurship ecosystem - the resource is not the problem, the resource mapping is

What it does

  • Our AI-powered chatbot, OskiGPT, acts as a personal guide, streamlining access to entrepreneurial resources at UC Berkeley. It's designed to save students hours of research time by providing tailored recommendations based on their individual needs. Whether a founder is seeking funding resources, mentorship programs, or legal advice, OskiGPT has them covered.

How we built it

Develop our backend environment using GPT-3.5 Web Crawling API.

  • This API allows us to return a complete affiliate list of URLs associated with UC Berkeley startup organizations, parse text information from these webpages, and feed the information to GPT model for machine learning.

Develop our frontend as a mobile App.

  • We connected our backend algorithms with our frontend, allowing future users to interact with our solution through Q&A conversations.

Design and implement mobile app UI.

  • We designed a user-friendly mobile app UI that provides future users an easy-to-use interface to ask questions and obtain accurate answers instantly.

QA Test MVP with three website crawled information

Challenges we ran into

  • Due to the limited number of tokens we can use per request is limited, our chatbot can only search within the limited number of rows of csv data, which results in limited answer, especially for complex questions.
  • Since we have to crawl the numerous websites with multiple layers of web pages, our csv data files are huge and it takes long time to extract the data for embedding.
  • There are several websites do not allow web crawling which results bugs (We would need to work with the organizations of the websites later to solve the problem)

Accomplishments that we're proud of

  • We built our first MVP from scratch within one day working closely together.
  • We quickly learned how to modify codes and integrating Open AI API effectively.
  • We learned each other's talents and expertise quickly, splitting specialty work among team members to progress on our product roadmap with lightning speed.
  • We learned how to design and prototype in Figma with zero previous experience.

What we learned

  • We learned tremendously about FeatureForm, LlamaIndex, and web crawling Q&A API from Open AI.
  • We gained a deeper appreciation on the challenge many startup founders have been facing on locating the information they need. The customer empathy we cultivated from this Hackathon helps us consider more deeply on building our solution more user-centrically.
  • Formed and strengthened friendship and allyship among team members. We are so amazed by the diverse range of expertise and talents each team member possesses through this Hackathon.

What's next for Oski GPT

  • We plan to continue refining the AI algorithms, improving its precision over time. We will request for lifting the rate limit in order to web crawl more UC Berkeley startup/entrepreneurship websites.
  • The chatbot service will initially be offered for free, and later, we plan to introduce premium features such as personalized mentor matching and advanced industry research to generate revenue.
  • We will also consider expanding the scope of data to include UC Berkeley resources outside of entrepreneurship or(and) other UC schools' resources.

Built With

Share this project:

Updates