Inspiration
Restaurants that some of us had worked at had a problem with calls being missed due to business in the restaurant itself. Given that much business is lost at restaurants to these situations, we attempted to make a product for them that would continue bringing in customers without being too harsh technically.
What it does
Auxilio allows restaurants to connect to an intelligent GPT3 chatbot that allows for it to place orders and reservations without having to specify some sort of strict schema, as shots have already been generated and the restaurant can add their menu to the site by performing an API request. Unfortunately, the project is still a bit technical in that API calls are required to add their menu to the application which will ensure the responses are in check
How we built it
We used mongodb, javascript, twilio and various other js libraries to create an interface between texting on the phone and GPT to generate a response based on the prompt from the text to allow flexibility in how a person replied yet still being accurate:
Challenges we ran into
The biggest challenge we ran into was prompt engineering, in that we were confused in how verbose to be or how vauge we should've been, and this ended up costing us a lot of our time for the project overall.
Accomplishments that we're proud of
Once we figured out how to complete the prompt engineering however, we were amazing in how it felt like we were programming in english and how remarkbly simple our shots had to be. The library that is used is also very special in that it is an extension upon GPT3 in that it introduces a context into the token space, which one of our team members had worked on separately.
What we learned
We learned how to use new apis that we weren't familiar with along with interfacing with GPT3 that can be used to inform other projects in the future in which the output space can't be well defined by some hardcoded schema, hence calling on GPT3's flexibility
What's next for Auxilium
Designing a proper frontend that will allow a restaurant owner to view upcoming owners (right now it is extremely rudimentary, just a page for the json of all customer outputs owners would have to hookup)
Log in or sign up for Devpost to join the conversation.