CanineIQ - Are You Actually Ready for the Dog You Want?

This is a submission for Weekend Challenge: Dog Days Edition

What I Built

I watched the CCTV footage in this India Today report of a couple being mauled by a pitbull while they viewed a rental home. The moment they opened the gate, the dog was on them - and it kept attacking even after its owner ordered it to stop, because it had been left to roam free. The dog wasn't evil. The ownership was irresponsible.

That video is why CanineIQ exists. It's a dog welfare intervention disguised as an assessment - 8 blunt questions about living situation, experience, time, training plan, budget, household safety, motivation, and how stable your life is. Then it scores how well your answers match the documented needs of the breed you want. Not stereotypes. Not marketing. The actual work the dog will need from you.

You walk away with three things most "which dog should I get?" quizzes refuse to give you:

1. A deterministic 0–100 readiness score with a plain verdict: READY, CAUTION, or NOT_READY.
2. A letter written from the dog's perspective, read aloud in a soft, sympathetic voice - it calls you out if you chose the breed for status, and tells you what it actually needs if you're not ready.
3. A breed health transparency page documenting the genetic damage we've bred into popular dogs - French Bulldogs needing palate surgery by age 3, Pugs that can't breathe during exercise - with sources.

The name says it: CanineIQ - Are You Ready? It's built for International Dog Day because every year thousands of powerful "status breeds" get surrendered, neglected, or end up in bite incidents because the humans who chose them weren't ready for the responsibility.

The other half of the same problem: French Bulldogs are the world's most popular breed, and a huge share of them will need palate surgery by age three because we bred them to have faces that can't function. The dogs deserve honesty even when the humans buying them don't want to hear it.

Demo

Live: https://canineiq.vercel.app/
Video: https://youtu.be/C8REqujvyx4

It works out of the box - no API keys to set up, no desktop required. The whole flow (assessment, verdict, the dog's letter, read-aloud voice) is tested and usable on a real Android phone as well as desktop.

Try the worst-case: pick a Great Dane, answer every question with the least responsible option (no budget, no experience, small apartment, young kids, "I want it for protection"), and listen to the letter the dog writes you. Then try the French Bulldog on the health page and read what its breathing is really like.

Code

• Repository: https://github.com/AnubhavDash/CanineIQ

The repo contains the full React SPA, the Vercel serverless functions (api/), the deterministic scoring engine, the 40-breed trait table, and the fact-checked health content with sources.

How I Built It

The scoring is the hard part, and it's not an LLM

The core design decision: Gemini never decides the score. A deterministic engine (api/score.js) weighs each of the 8 answers against the chosen breed's documented needs - energy, space, fragility, grooming, health cost, household safety, trainability, difficulty. That produces a 0–100 score and the verdict (READY ≥ 70, CAUTION ≥ 40, NOT_READY below).

Gemini's job is writing: a personalised narrative built around that fixed score, with top warnings, genuine strengths, per-answer findings, and the dog's letter. The API validates and normalizes every response server-side, so a hallucinated score or an off-script alternate-breed suggestion gets overridden deterministically before the user ever sees it. The LLM can't lie to you; that's the whole point of a welfare tool.

Rate-limit resilience

Both the assessment and photo-identification functions run a model fallback chain (gemini-3.5-flash-lite → gemini-3.1-flash-lite → gemini-3.6-flash), so a popular demo can't kill the app. Same for the speech endpoint (eleven_v3 → eleven_multilingual_v2 → eleven_flash_v2_5), and the browser falls back to native SpeechSynthesis if the API is unavailable.

40 breeds, 10 photographed locally, 30 lazily loaded

Ten core breeds ship as local assets; the other 30 load images from the Dog CEO API via a searchable dropdown. You can also snap a photo of a dog and Gemini's multimodal model identifies the breed, auto-selecting it for the assessment.

The dog speaks

The signature feature is a letter written from the dog's point of view, generated alongside the verdict and read aloud with ElevenLabs emotional text-to-speech - a soft, sympathetic voice with [sighs] delivery. If you chose the breed for status, it calls that out. The verdict strip is unmissable: breed, score, verdict, right above the letter.

Honest health content

The Breed Health page documents what was done to these breeds genetically, what the dog experiences day-to-day, surgery rates, what breeders won't tell buyers, and whether it's ethical to buy one at all - citing OFA, AKC Canine Health Foundation, The Kennel Club (UK), RSPCA, and Cambridge/Cornell veterinary research. Figures are framed as qualitative ranges because inherited-disease risk is line-specific, not a single statistic.

Prize Categories

• Best Use of Google AI - the assessment narrative engine, breed health analysis, and photo-based breed identification all run on the Gemini API, with strict server-side validation so the model never controls the score.
• Best Use of ElevenLabs - the dog's letter is spoken in an emotional voice profile, with a graceful fallback to browser speech synthesis.

Comments

[Anubhav Dash]

🐕 CanineIQ — Are you actually ready for the dog you want? An 8-question readiness check that scores your life against the breed's real needs. The dog writes you a letter, read aloud.

🔗 Live: canineiq.vercel.app/
🎬 Video: youtu.be/C8REqujvyx4
🐙 Repo: github.com/AnubhavDash/CanineIQ