Inspiration
We were inspired by how often we see organizations with systemic blind spots. If a decision maker doesn’t know they should search for something, even the best search can’t help.
What it does
Blindspot starts with your knowledge and continually explores. By conversing about your knowledge, Blindspot saves what you do know and tells you what you don’t.
How we built it
We used a vector database to store your knowledge, and continually make queries as the user converses with the LLM. We mark all of the explored knowledge so we can shift the conversation to unexplored knowledge. We used react for the frontend.
Challenges we ran into
We needed to guide the LLM towards unexplored territory. We eventually settled on not reusing the documents already sent to the LLM/user.
Accomplishments that we're proud of
We used building a personal fitness plan as a test case. We’re proud to say GPT4 gave us advice on sleep quality.
What we learned
GPT4 tends to be fairly creative, we had to make it more focused.
What's next for Blindspot
We’d like to index more documents both internal and external to the organization. This should help Blindspot get better at giving advice.
Log in or sign up for Devpost to join the conversation.