This is a submission for Weekend Challenge: Dog Days Edition
What I Built
I built Dog Roaster - a small web app deployed on vercel. You can upload an image of your dog (format - jpg, png, max size - 8MB), and the Gemini AI (model: 3.6-flash) would generate a roast line.
The idea is simple on purpose: one upload, one API call, one roast. No sign-up, no clutter. Just upload, or drag and drop, get humbled.
Demo
Link: https://dog-roast.vercel.app/
Code
Github link: https://github.com/MIRA-KUMAR/dog-roast
README.md has instructions to run to locally and modify the model and prompt accordingly.
How I Built It
The app is a lightweight frontend + serverless backend, deployed on Vercel:
Image upload — users upload a dog photo directly from the browser (jpg/png, capped at 8MB to keep things fast and within API limits).
Gemini Vision — the image is sent to Google's @google/genai SDK using gemini-3.6-flash, prompted to look at the photo and generate a short, punchy roast rather than a neutral description.
Retry handling — API calls are wrapped in a retry function to gracefully handle transient 503 errors from the model.
Serverless deployment — the whole thing runs on Vercel.
The trickiest part was actually picking the right model. I started with gemini-3.7-flash since it's the newest, but it kept throwing 503s under high traffic. Switching to gemini-3.6-flash fixed it immediately: still fast, still sharp on the roasts.
Prize Categories
Submitting to Best Use of Google AI — Gemini, using the image upload directly as model input and outputting text.
Top comments (0)