C
AI/ML
Web
Hackathon

Le Pathétique (BearHacks 2026 — AI French Food Critic)

Solo BearHacks 2026 build. Upload a photo of your sad cooking, get roasted out loud by a pretentious French AI critic, then redeemed with broke / decent / fancy recipes.

Confidentiality note

A Vercel deployment exists but has known issues being polished post-hackathon — DevPost link is the source of truth for now.

Live demoGitHubBack to projects

Game stats

Hours on record

48 hrs

Last played

Last weekend

Achievements

13 / 13

Reviews

Overwhelmingly Positive (88)

Install size

78 MB

Players

Single + Co-op

Media

Achievements

4 of 4 unlocked · 13 of 13 per Steam

  • Solo Hack Shipped

    Built, shipped, and demoed Le Pathétique at BearHacks 2026 — solo, in one weekend.

  • Multimodal in One Shot

    Gemini Flash handled both vision + creative writing in a single call. No chaining required.

  • Wi-Fi Will Not Stop Me

    Built an offline mock mode so the demo survives bad venue Wi-Fi or revoked keys.

  • 13 Easter Eggs

    Six critic personas, thirteen hidden easter eggs. Repeat-play tested.

Overview

Problem

Hackathon prompt: ship something that uses multimodal AI in a way that's actually fun, not a demo with three buttons. Needed an interaction that survives a shaky Wi-Fi judging room and doesn't depend on warm API caches.

My role

Solo developer. Designed the persona, built the pipeline (Vision → Gemini → ElevenLabs), wrote the brutalist newspaper UI, wired the offline mock mode for demo resilience, hid 13 easter eggs across 6 personas.

Stack

Next.js 16 (App Router) • React 19 • TypeScript • Tailwind v4 • Google Cloud Vision API (image labeling) • Gemini Flash (multimodal critique + recipe generation) • ElevenLabs (text-to-speech, French accent) • Web Audio API (in-browser SFX) • Vercel (hosting)

Highlights

  • Shipped solo at BearHacks 2026 — concept to working multimodal demo in one weekend.
  • Built an offline mock mode so the app keeps working when the venue Wi-Fi or an API key dies mid-judging.
  • Designed 6 distinct critic personas with 13 hidden easter eggs to reward repeat use.
  • Brutalist newspaper aesthetic that stood out from the standard hackathon Tailwind look.

Case study

Le Pathétique (BearHacks 2026)

DevPost: https://devpost.com/software/le-pathetique GitHub: https://github.com/Cengizbey-m/Le-Pathetique

A pretentious AI French food critic that roasts your cooking out loud, then redeems you with three budget-tiered recipes. Built solo at BearHacks 2026 in one weekend.

The pitch

You upload a photo of whatever sad meal you're staring at. The app analyzes it, then a French-accented voice critiques it like Gordon Ramsay's snobbier cousin — that tone hovering between disappointment and fury. Then it offers redemption: three recipes tagged broke, decent, and fancy so you can fix dinner at whatever budget you've got.

The pipeline

  1. Photo upload in the browser, no server upload yet.
  2. Google Cloud Vision API labels what's actually in the picture (so the AI doesn't hallucinate "bouillabaisse" when you're holding a granola bar).
  3. Gemini Flash takes the labels + the photo and writes the critique + three recipes in one multimodal pass.
  4. ElevenLabs synthesizes the critique with a French-accented voice.
  5. Web Audio API layers in in-browser SFX (no asset downloads).

What I'm proud of

  • Offline mock mode. Hackathon Wi-Fi is a coin flip. If Vision or Gemini fails (or a key gets revoked at 2 AM), the app falls back to canned-but-plausible critiques and recipes. The demo never goes blank on stage.
  • Six personas, 13 easter eggs. Repeat users find different critics with different voices and small hidden interactions. Built for replay value.
  • Brutalist newspaper aesthetic. Most hackathon UIs blur together — Tailwind defaults, soft shadows, gradient buttons. I went the opposite way: serif headlines, harsh borders, monochrome with one ink-red accent. Stood out on the judging table.
  • Multimodal in one shot. Gemini handles both the visual analysis and the creative writing in a single call, instead of chaining a separate vision model and an LLM.

Tech

  • Next.js 16 (App Router), React 19, TypeScript, Tailwind v4
  • Google Cloud Vision API — image labeling
  • Gemini Flash — multimodal critique + recipe generation
  • ElevenLabs — text-to-speech with French accent
  • Web Audio API — sound effects, no audio assets
  • Vercel — hosting

Status

  • Hackathon: BearHacks 2026 · solo build
  • Vercel deployment exists but has known issues post-hackathon. DevPost link is the source of truth for now.
  • Source is open: github.com/Cengizbey-m/Le-Pathetique.