**Roast Or Praise** - AI-Powered Setup Critiques with Groq Vision
Upload a photo of your desk setup, room, or objects and get playfully roasted or praised by AI with a detailed 0-10 score and witty feedback. Built with Groq’s vision models, it generates shareable report cards for social media.
I’ve been heavily using Cursor to tackle a lot of the base code and generating the agent prompts. This felt like a fun idea to use vision models for - they’re incredibly fast, have a generous free tier to get started, and support structured outputs. This meant I could build an app that’s both fun and actually usable, not just a demo. The “Praise” mode was added because sometimes you want validation instead of a roast, making it more versatile and inclusive. It is using Groq’s native JSON schema enforcement for structured output which is very nice. When you decide to share your roast/praise it runs the picture through another Groq request that does a bit of image moderation, if its obvious offensive or illegal it will get flagged and it will be masked with a blur.
On the roast/praise Groq will also supply me with something that would enhance your setup or room or whatever you decide to send in. That will be used to create a direct Amazon associates search link based on said item 
For a office setup it could be a cable management box or a plastic plant.
Some things I’d like to work on a bit more is a better self-moderation system and blurring, and also try out the SDK instead of my own API fetches. And obviously my biggest weakness - the design
