Confidant
Confidant is a web-based tool that uses the power of large language models to assist with solving secure multi-party computational problems.
Web application
AI / LLM
React
Email authentication
What got us started
Two millionaires want to find out who’s richer, without telling each other how much they each have got in the bank.
This is an example for Yao’s millionaire problem, or a ’secure multi-party computational problem‘, to put it more formal.
We attempt to solve this with the help of artificial intelligence.
How it works
Users provide a question (to be answered by themself and others) and an objective (to be determined by the tool). Participants are invited to submit their answers via email.
Once all answers have been collected, Confidant establishes the objective, sends out the result and deletes the data. With none of the participants ever knowing what exactly the others have answered.
The technical side
In order to make Confidant easily accessible for everyone, we utilized React to create a web application that can be accessed from any browser. User session data is collected via a form and temporarily saved.
Emails are based on simple HTML templates and automatically populated with user data before being send out.
Once all answers have been collected, they are sent to OpenAI via their API for ChatGPT to evaluate the result. In addition to the user data, we also send a specifically created prompt that instructs the model what to do.
After the result has been sent out to users, all data from the session is being deleted. In case not all participants answer within 30 days (and therefore no result can be determined) the session data is automatically deleted and all users are informed of the termination.