What I Built PawTalk give your dog a voice.
Upload a photo, answer a few quick questions, and AI sketches out a playful personality for your dog: traits, energy level, a secret talent, a biggest obsession.
Then you can actually talk to them ask a question, get an in-character reply, and hear it spoken aloud in their own voice.
There's also a bark translator: record your real dog barking, and it gets "translated" into one funny in-character line.
Everything is framed honestly as playful AI interpretation, never a real claim about what a dog is thinking — no fake science, just a fun way to spend time with your best friend.
Demo https://pawtalk-murex.vercel.app/ Code https://github.com/Olobaray20/pawtalk How I Built It Next.js 14 (App Router) + TypeScript + Tailwind, with three server-only API routes doing the real work: Google Gemini () handles vision reading the uploaded photo alongside the facts you typed in and generates the structured personality, the in-character chat replies, and the bark "translations." The personality is stored as one structured object and re-sent as context on every chat turn, so the dog's voice stays consistent instead of drifting message to message.
ElevenLabs turns every reply into actual speech, auto-played the moment it arrives so hearing the dog is the default experience, not an extra click.
The bark translator records a few seconds of mic audio and sends it to Gemini as inline audio data, the same mechanism as the photo one dog, one AI pipeline, three input types (image, text, audio).
The one design decision I'm proudest of: every AI call has a demo-mode fallback.
No API key, a timeout, a rate limit, a safety filter none of it crashes the app.
It falls back to offline canned content and shows a clear "Live AI" vs "Demo mode" badge everywhere, so the product is honest about what's real and never breaks mid-demo.
Prize Categories Best Use of ElevenLabs, Best Use of Google AI