Inspiration
We were inspired by the Marshall Wace chain of event project, and thought applying that concept to recipes would be a natural way of fitting it, given cooking is just a chain of events towards an end meal.
What it does
Takes as input an image from a file or a camera, produces an ingredients list from what is visible in the image, then produces a step by step recipe until a plausible end meal is created visualising this process as a chain of events.
How we built it
We used the OpenAI API to handle our AI data processing tasks, coding the endpoints in python using constrained generation to ensure the LLM outputs conformed to a desired JSON schema, while prompt engineering to ensure its as consistent as possible in outputting what we want. We hosted the server with a Flask backend and used a react frontend.
Challenges we ran into
Integrating our backend into our frontend has proven very challenging, mainly due to the lack of frontend experience on the team making debugging take a while.
Accomplishments that we're proud of
Making our AI system work as intended quite quickly, and overcoming many challenges when integrating the whole system together.
What we learned
Keep the frontend simple when time is constrained, over-engineering will bite you
What's next for Gastrochain
Continuing to add more functionality that we envisioned throughout the project to provide richer event interpretations and more accurate recipes.
Log in or sign up for Devpost to join the conversation.