Developed my AI-Powered Meal Generator
techI have been using openAI's chatGPT to compliment on my day to day activities and routine to optimise my work and decided to explore and be on the AI side which is integration.
The capabilities of AI has been overwhelming and has taken tech space to a storm so I decided to start with a simple Meal Generator - meet here.
Creation of Buzz Eats
As a fitness and health enthusiast, one of my daily struggles is deciding on what to eat whether I want to be strict with my calorie intake or I just want to simply EAT and on top of that I'm a techie who also wants to get my hands on AI so I created Buzz Eats.
Buzz Eats gives me simple and quick meals based of whether I am following a calorie intake, allergens, diet restrictions, or whatever I have in my kitchen. I initially just wanted a random meal. Apparently, I realized I don't have majority of the ingredients AI is giving me. I ended up adding a field to be able to type in the ingredients I have. In a few rounds of refinement, I realized I'm too lazy to type and would just want to take a photo of what I want to make a meal of. Whilst ingredients is not required, Buzz Eats still can generate random meals based from user's preference.
Buzz Eats at core
Buzz Eats is an AI-powered meal generator using OpenAI's GPT-4o-mini. Its prompt has been refined and configured to accept a combination of image and text to make sure it gives the most accurate response out from the data. I am still fine-tuning Buzz Eats such as making sure it recognizes digestible objects only from the inputs from text and image uploaded. It can evaluate when the data provided are in conflict and should still be able to decide which takes precedence to get the the most appropriate response.
Buzz Eats is still in its early experimental stage so it may break from time to time and new features are on its way as well. If you had a chance to play around whether BuzzEats was nice or nasty - or you think it needs some spiffing up, I would love to hear your thoughts! :)
Beginner's Takeaway
Developing the most efficient prompt takes a lot of refinement and fine-tuning. OpenAI was learning and as the Prompt Engineer for Buzz Eats, I was learning to adjust with every response OpenAI was giving me making sure the context are as detailed and precise to get the best response and avoid hallucination.
Buzz Eats behind the screen
Buzz Eats is built with Next.js, TailwindCss together with shadcn/ui in the frontend. The backend is built using Supabase's Edge Function and I have utilised the Supabase's Storage to store images temporarily before it goes to OpenAI. You can play with BuzzEats here or check out the code here.