DEV Community: Tarun ★

MenuMind Arcade 🍽️🎮 — AI Smart Menus & Gamified Food Quests

Tarun ★ — Thu, 15 Oct 2026 05:56:28 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

MenuMind Arcade 🍽️🎮 — AI Smart Menus & Gamified Food Quests for Independent Eateries

What I Built

The Elevator Pitch

MenuMind Arcade transforms boring, static PDF QR menus into dynamic, gamified culinary discovery experiences powered by Google Gemma 3 running locally.

Independent restaurants, boutique cafés, food trucks, and specialty roasteries publish colorful interactive menus complete with Indian Rupee (₹) pricing, spice gauges (0–4 🌶️), and verified kitchen allergens. Diners simply scan a table QR code and embark on a 7-step Food Quest — an intuitive quiz that matches their exact budget, hunger level, spice tolerance, flavor cravings, and strict dietary requirements against the live kitchen inventory.

Who Would Pay For It?

Our primary customers are independent restaurateurs, café operators, and food truck owners:

Boutique Cafés & Artisan Roasteries with rapidly evolving seasonal harvests and craft beverages.
Dietary-First Eateries (100% Vegan, Jain-friendly, Gluten-Free) where guests spend 10+ minutes quizzing busy waitstaff before feeling safe to order.
High-Volume Food Trucks & Pop-up Kiosks operating with lean crews that need fast, confident self-discovery menus.

The Problem It Solves

The Dreaded "Pinch-and-Zoom" PDF: Cardboard table stands directing customers to 15MB PDF scans create frustrating mobile experiences with zero searchability, unreadable fine print, and no guidance on portion sizes or heat levels.
Allergen Anxiety & Kitchen Confusion: Cross-contact concerns and dietary requirements (Jain, Vegan, Nut allergies) cause dining friction, waitstaff bottlenecks, and accidental food returns.
The AI Wrapper SaaS Trap: Enterprise digital menu providers charge upwards of $150–$300/month. Meanwhile, naive AI wrappers that funnel every menu query to proprietary cloud APIs (like GPT-4 or Claude) rack up unpredictably high per-token invoices that wipe out thin restaurant margins.

The Side Hustle Monetization Model

MenuMind Arcade operates on a transparent, high-margin, dual-tier subscription:

Tier	Price	Highlights
Starter Community	₹0 (Free Forever)	1 Active Restaurant, up to 25 dishes, standard QR code, deterministic rule-based matchmaker, full allergen safety.
Arcade Pro	₹799 / month (~$9.50 USD)	Unlimited menu items, Google Gemma 3 local AI sommelier reasoning, AI Text-to-Menu bulk importer, high-res branded table tent QR generator, and real-time customer craving analytics.

Because MenuMind Arcade runs inference locally using open-weight models, our marginal cost per customer query is ₹0.00, making a ₹799/mo subscription an incredible 90%+ gross-margin side hustle!

Demo

Video:https://youtu.be/98RD17mms-g
Github :(https://github.com/TarunTXE/menumind-arcade

Deterministic Guardrails First: Before touching an AI model, the server filters the entire menu using hard constraints:
- Dish price / customer budget in INR.
- Required dietary tags (Vegan, Jain, Gluten-Free) must be present.
- Avoided allergens (Dairy, Nuts, Soy) are strictly excluded using negative lookahead/lookbehind regex matching, ensuring tags like dairy-free are never falsely discarded.
- Out-of-stock items are automatically excluded.
Gemma 3 Soft-Constraint Reasoning: Only sanitized, pre-filtered candidates are passed to gemma3:4b. Gemma acts as a seasoned sommelier, evaluating subtle craving harmonies, spice pairing, and hunger satisfaction.
Multi-Tier Model Fallback: If gemma3:4b takes longer than 25 seconds or local GPU resources are constrained, the orchestrator cascades to gemma3:1b (0.3s response time), and finally to our deterministic rule-based matchmaker with full provider attribution transparency.

Why Does Open Innovation Matter?

1. Economics: Destroying the "Per-Token Tax" on Small Business

A bustling neighborhood café serves 300 to 600 diners per day. If each table scans the QR code and runs 2 or 3 Food Quests, a closed API wrapper charging $0.02 per request incurs $180 to $360 per month in raw API fees. For an independent eatery running on 8–12% net margins, that cost is prohibitive.

By leveraging open-weight Gemma 3 models via Ollama, inference runs entirely on the host machine or a shared edge node for $0.00 in per-token bills. Open innovation enables sustainable side hustles without cloud API anxiety.

2. Guest Privacy & Data Sovereignty

Dietary preferences, medical allergen disclosures, and dining habits are deeply personal. With closed cloud APIs, every customer craving and dietary restriction is transmitted to third-party proprietary servers. With local open-weight inference, dining preferences remain strictly on local infrastructure, protecting customer privacy.

3. Offline Resilience in Fast-Paced Kitchens

Hospitality venues frequently experience WiFi drops, spotty cellular connections, and ISP disruptions. Cloud-only menus go blank when the internet fails. Because MenuMind Arcade pairs local Gemma inference with an isolated in-memory/embedded data store and deterministic fallback, the digital menu and discovery engine keep running even during internet blackouts.

4. Gemma Licensing & Open Governance

Google releases Gemma under the Gemma Terms of Use — an open-weight commercial-friendly license. While not an unrestricted OSI license, it allows commercial side hustles to build, deploy, and monetize intelligent applications freely worldwide, while upholding safety and ethical use standards.

My First User

During development, I shared the prototype with Priyanshu, co-founder of an artisanal plant-based roastery and brunch bistro (Green Fork Café prototype):

*"Our weekend brunch rush is usually chaotic. Customers constantly ask whether the pesto contains parmesan or if the cold brew tonic is strictly vegan. We put the MenuMind Arcade QR code on three patio tables during Sunday brunch.

Guests loved the Food Quest quiz — they treated it like a mini game on their phone. The ₹450 budget slider gave them price confidence, and nobody had to wave down a waiter just to double-check allergens. If this costs ₹799 a month with the table QR generator and menu text importer, I'm signing up on day one."*

Priyanshu's feedback directly inspired our Verified Kitchen Allergen Policy badge, the instant availability toggle on the owner dashboard, and the 1-click "View on Menu" button on quest recommendations!

Prize Categories

We are entering MenuMind Arcade into the following tracks for the Hacktoberfest Open-Source AI Challenge Week 2:

🏆 Primary Challenge: Week 2 — Side Hustle A viable, production-ready, monetizable business model for independent restaurants and food trucks powered by open-source AI with ₹0 marginal token costs.
🤖 Partner Track: Google Gemma / Local Open-Weight Inference Showcasing real local execution of Google Gemma 3 (gemma3:4b and gemma3:1b) through Ollama, paired with deterministic safety guardrails.

BugDex - Gotta find em all

Tarun ★ — Wed, 07 Oct 2026 00:34:27 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

🌿 What I Built

BugDex is an AI-powered real-world insect collection game and digital field journal designed to turn screen time into an outdoor adventure. Inspired by classic monster-catching compendiums like the Pokédex, BugDex empowers people to step outside, explore backyards, parks, campus gardens, and hiking trails, and discover the fascinating mini-beasts living all around us.

Whenever you spot an insect, spider, or arthropod in the wild:

Snap a photo using your smartphone or upload a picture.
AI Multimodal Vision identifies the specimen in real-time, extracting its biological taxonomy (Order, Family, Scientific Name, and Common Name), morphological traits (wing venation, elytra, antennae), habitat, diet, and human safety guidance.
Claim a Holographic Collectible Card: Each verified observation transforms into a collectible card ranked across rarity tiers (Common, Uncommon, Rare, Very Rare, and Legendary) with animated foil sheens, confidence scores, and ecological notes.
Level Up & Embark on Daily Outdoor Quests: Earn XP, unlock ranks (from Novice Bug Catcher to Master Entomologist), maintain daily exploration streaks, and complete daily expeditions verified directly by Vision AI (e.g., "Find an insect on a flower", "Spot a winged flyer", "Find a micro-marvel smaller than a fingernail").

🎯 Who is it for?

Outdoor Explorers & Nature Lovers seeking an interactive field guide to understand local biodiversity.
Kids, Families & Students who want to learn biological taxonomy through hands-on gamification rather than dry textbooks.
Developers & Tech Enthusiasts who need an engaging, rewarding excuse to close their IDEs, take a walk, and truly "touch grass".

🎥 Demo

Demo Video / Walkthrough: https://youtu.be/nTn3YaPzTEk

✨ Key Features at a Glance

📸 Instant Multimodal AI Field Identification: Sub-second visual classification with real-world biological accuracy.
🎴 Holographic Foil Specimen Cards: Dynamic 3D tilt effects, animated rarity gleams, and comprehensive ecological field notes.
📜 Interactive BugDex & Journal: Filter by taxonomic order (Coleoptera, Lepidoptera, Hymenoptera, Mantodea, Odonata, etc.) and rarity with smart deduplication and sighting counters.
🎯 AI-Verified Daily Expeditions: Daily outdoor quest verification where AI assesses whether your real-world snapshot fulfills the day's biodiversity mission.
🏆 Achievements & Leveling System: Earn badges (🐛 First Bug, 🦋 Collector, 🌙 Night Hunter, 🔥 Streak Adventurer) and progress across 6 naturalist ranks.
⚠️ Wildlife Safety First: Automatic warnings for stinging or venomous species paired with advice to observe respectfully and never harm specimens.
🔒 Privacy-First & Offline-Ready: Client-side storage with broad habitat tagging (Park, Garden, Trail)—no intrusive GPS tracking.

💻 Code

The entire project is open source and hosted on GitHub:
🔗 Repository: https://github.com/TarunTXE/BugDex

🛠️ How I Built It

BugDex is built around a modern, reactive, mobile-first web architecture combined with multimodal vision models:

🧠 Open-Source & Multimodal AI Architecture

BugDex utilizes the Google Gemini Multimodal Vision REST API (@google/genai), implementing an ultra-low-latency model hierarchy with resilient fallback handling:

gemini-flash-lite-latest (Primary for instant outdoor scanning)
gemini-3.5-flash-lite
gemini-3.1-flash-lite
gemini-2.5-flash-lite
gemini-3.6-flash (Deep analysis fallback)

🔍 Strict Taxonomy & Vision Pipeline

To prevent hallucinations and guarantee scientific fidelity:

Base64 Inline Buffer Ingestion: Live camera captures or user-selected images are downscaled and encoded client-side to minimize mobile data usage.
Constrained JSON Schema Output: AI prompts enforce structured biological schemas (response_mime_type: "application/json") with low temperature (0.1), extracting exact scientific binomial nomenclature, family, order, confidence rating, appearance breakdown, diet, and human safety instructions.
Negative Sample & Uncertainty Rejection: If an uploaded photo contains human faces, pets, or non-living objects, the model gracefully returns "is_insect": false, guiding the user with framing tips.
AI Quest Evaluation Engine: A dedicated prompt engine evaluates photos against active daily challenges, analyzing the scene context (e.g. insect perched on a petal vs flying in midair) and returning custom verification feedback.

🎨 Frontend & Ergonomics

React 19 & Vite: Ultra-fast hot-reloading and lightning-fast bundle performance.
Tailwind CSS: Custom dark nature glassmorphism, glowing taxonomic badges, and CSS keyframe holographic foil sheen animations.
Web Audio API Sound Synthesizer: Custom synthesized retro field sound effects for scanning, level-ups, and card unlocks without external audio asset bloat.
Canvas-Confetti: Visual celebration bursts upon completing quests or uncovering rare specimens.

🌍 Why Does Open Innovation Matter?

Open innovation in AI and scientific education is a transformative force:

Democratizing Biodiversity Education: Traditional field guides and taxonomic keys are often dense, academic, and inaccessible to casual nature enthusiasts. Open AI models turn complex biological taxonomy into an accessible, interactive digital companion in the palm of anyone's hand.
Citizen Science at Scale: When biodiversity tools are open, extensible, and free from proprietary lock-in, anyone can contribute to local ecological awareness, documenting urban pollinators and tracking seasonal insect populations in their local communities.
Transparency & Safety: Because nature exploration involves real creatures—some of which can sting or bite—open, reproducible prompt architectures ensure that safety guidelines, non-invasive observation practices, and data privacy remain transparent and community-audited.
Connecting Technology with the Living Earth: Open innovation reminds us that technology should not isolate us in virtual walled gardens; instead, it can serve as a lens that deepens our awe, curiosity, and respect for the natural ecosystems right outside our front door.

🏆 Prize Categories

Hacktoberfest Week 1: Touch Grass (Main Track)
Open-Source AI & Vision Innovation

Go outside, explore your local greenery, and happy bug hunting! 🐛🌿🔍