Sensory Culinary Intelligence & Mood-Based Food Recommendation System
Explainable, constraint-aware psychological emotional regulation through dietary intervention.
Repository • Architecture • Benchmarks • Getting Started • Author & License
MoodBite.Ai is not a chatbot wrapper.
While conversational tools answer general trivia, MoodBite.Ai is a dedicated, explainable, domain-specific culinary intelligence system designed to regulate human psychological and physiological states through food and beverage recommendations:
- No Smart Fridge Required: MoodBite does not require an expensive physical smart refrigerator. It recommends meals, beverages, and snacks accessible via home cooking, local markets, nearby diners, or food delivery platforms.
- 100% Free & Zero Paid APIs: No subscriptions to paid OpenAI, Gemini, or third-party vector databases are needed. The entire recommendation engine executes deterministically and locally in
<3 ms. - Zero Constraint Hallucinations: Strict hard constraints (Vegetarian, Vegan, Non-Veg, Make at Home, Order Online) are deterministically enforced before scoring. A vegetarian user will never be recommended a meat dish.
- Explainable Rationale: Every recommendation provides a scientific Chef's Note detailing how specific nutrients (tryptophan, serotonin, complex carbohydrates, magnesium, capsaicin) counterbalance the detected emotional vector.
MoodBite runs on an Emotional Vector Machine (EVM) hybrid architecture with a 12-layer deterministic pipeline:
flowchart TD
A["👤 User Input (Text / Voice / Ingredients)"] --> B["Layer 1: Intent & Guardrail Gate"]
B -->|Greeting / Off-Topic / Gibberish| C["Guided Clarification & Prompts"]
B -->|Valid Culinary / Emotional Query| D["Layer 2: Clause & Negation Engine"]
D -->|Contrastive Clauses & Scopes| E["Layer 3: Canonical Emotion Ontology"]
E --> F["Layer 4: EVM Vector Fusion (Valence, Arousal, Control)"]
F --> G["Layer 5: Calibrated Confidence Estimator"]
G -->|Low Confidence / Ambiguous| H["Disambiguation Clarification State"]
G -->|High Confidence| I["Layer 6: Sensory Strategy Mapper"]
I --> J["Layer 7: Hard Constraint Deterministic Filter"]
J -->|Strict Pruning: Veg, Vegan, Non-Veg, Home/Delivery| K["Layer 8: Multi-Factor Scoring & Ranking Engine"]
K --> L["Layer 9: Adaptive Personalization & History Memory"]
L --> M["Layer 10: Explainable Chef's Note Generator"]
M --> N["Layer 11: High-Fidelity Verified Visual Catalog"]
N --> O["Layer 12: Atmospheric Fluid UI & PWA Output Dashboard"]
| Layer | Subsystem | Responsibility |
|---|---|---|
| 01 | Intent & Guardrail Gate | Classifies input into MOOD, FOOD_REQUEST, GREETING, OFF_TOPIC, or UNKNOWN. Prevents hallucinations for out-of-domain text. |
| 02 | Clause & Negation Engine | Splits complex sentences across contrastive conjunctions (but, however, although) and handles negation scopes ("not sad at all", "no longer stressed"). |
| 03 | Canonical Emotion Ontology | Maps linguistic signals into 11 canonical emotions (joy, sadness, anger, fear, stress, calm, excitement, boredom, loneliness, neutral, unknown). |
| 04 | EVM Vector Fusion | Computes continuous psychological coordinates across 3 dimensions: Valence |
| 05 | Calibrated Confidence | Assesses multi-expert agreement and signal separation to calculate true confidence (not arbitrary high numbers). |
| 06 | Strategy Mapper | Maps emotional vectors to target culinary interventions (e.g., Stress |
| 07 | Hard Constraint Filter | Deterministically eliminates invalid foods: strict Vegetarian, Vegan, Non-Veg, and availability constraints. 0% tolerance for violations. |
| 08 | Multi-Factor Ranking | Evaluates candidate foods using weighted scoring: emotional affinity, flavor profiles, preparation effort, ingredient overlap, and budget. |
| 09 | Adaptive Personalization | Tracks user feedback (thumbs up / thumbs down) in local storage, continuously tuning ingredient and cuisine weights without external trackers. |
| 10 | Explainable Chef's Note | Synthesizes an evidence-based explanation linking food chemistry to emotional state balance. |
| 11 | Verified Visual Catalog | Pairs recommendations with verified dish photography and reliable fallback rendering. |
| 12 | Atmospheric Fluid UI | Renders dynamic ambient gradients that adapt to the detected emotion, complete with PWA offline caching. |
MoodBite includes an offline evaluation suite (benchmark.js & tests/**/*.test.js) testing direct emotion queries, long-form natural language, contrastive negations, hard constraints, and edge cases:
| Metric | Score | Validation Standard |
|---|---|---|
| Intent Gate Accuracy | 100.0% | 14/14 test cases verified across all intent classes |
| Canonical Emotion Accuracy | 100.0% | 12/12 canonical emotion classes correctly identified |
| Negation Scoping Accuracy | 100.0% | Correct resolution of contrastive & double-negated sentences |
| Hard Constraint Safety | 100.0% | 0 violations across Vegetarian, Vegan, Non-Veg, Home, & Delivery |
| Disambiguation / Unknown Detection | 100.0% | Identifies gibberish and ambiguous queries without guessing |
| Average Pipeline Latency | < 2.8 ms | Real-time CPU inference with zero external network hops |
| Automated Unit Tests | 34 / 34 | Comprehensive Node.js native test runner test suite |
| External API Token Cost | $0.00 | 100% free and locally executable |
- 🎯 Precision Emotional Understanding: Accurately extracts emotional meaning from complex natural sentences like "My thesis review is tomorrow morning and I am unprepared" (mapped to Stress) or "I was feeling miserable earlier but things turned around!" (mapped to Joy).
- 🥗 Deterministic Dietary Enforcement:
- Vegetarian: Strictly excludes all meats, poultry, and fish.
- Vegan: Strictly excludes meat, dairy, eggs, and animal byproducts.
- Non-Veg: Prioritizes meat, poultry, seafood, or egg-based dishes.
- Make at Home / Order Online: Ranks dishes according to preparation feasibility and local availability.
- 💬 ChatGPT-Style History Drawer: Persistent session history with one-click restoration, mood tracking, and direct feedback submission.
- 💡 Adaptive Local Learning: Remembers what dishes you liked or disliked and refines future recommendations per user profile.
- 🌈 Dynamic Atmospheric Gradients: The landing page and output dashboard dynamically bathe in vibrant ambient glows and floating animated orbs that adapt in real time to the detected mood.
- 📱 Progressive Web App (PWA): Installable directly onto iOS, Android, macOS, and Windows with offline caching and standalone display.
MoodBite is fully optimized as an installable Progressive Web App:
-
Mobile (Android / Chrome):
- Tap the browser menu (
⋮)$\rightarrow$ select "Install App" or "Add to Home Screen".
- Tap the browser menu (
-
Mobile (iOS / Safari):
- Tap the Share button (
⎋)$\rightarrow$ scroll down and select "Add to Home Screen".
- Tap the Share button (
-
Desktop (Chrome / Edge / Brave):
- Click the Install icon on the right side of the URL address bar, or visit
/aboutto access the installation card.
- Click the Install icon on the right side of the URL address bar, or visit
- Node.js:
v18.17.0or higher (Node.jsv20+recommended) - npm:
v9.0.0or higher
git clone https://github.com/SinghCod3r/MoodBite.Ai.git
cd MoodBite.Ainpm installnpm run devOpen http://localhost:3000 in your web browser.
Run the full suite of 34 automated unit tests:
npm testExecute the comprehensive offline evaluation suite:
node benchmark.jsnpm run build
npm startMoodBite.Ai/
├── components/ # Modular UI components
│ ├── ChefsNote.js # Explainable recommendation rationale & alternatives
│ ├── ChatHistoryDrawer.js # ChatGPT-style session drawer & taste tracker
│ ├── HeroFoodCard.js # Hero dish card with verified photography & feedback
│ ├── ModernInputBar.js # Multi-modal input bar (text, speech recognition)
│ ├── MoodBackground.js # Atmospheric dynamic mood gradient & glowing orbs
│ ├── Navbar.js # Navigation bar with auth modal & history toggle
│ ├── PreferenceFilterBar.js # Strict dietary (Veg/Vegan/Non-Veg) & availability filters
│ ├── PWADownloadCard.js # Progressive Web App installation banner
│ └── UIStateView.js # Empty, clarification, ambiguous, and error views
├── context/ # React Context providers (Loading state)
├── evm/ # Core culinary intelligence & EVM vector engine
│ ├── dictionaries.js # Lexical definitions & emotion dictionaries
│ ├── emotionAnalyzer.js # Clause splitting, negation handling & emotion ontology
│ ├── experts.js # Multi-expert signal processors
│ ├── foodEngine.js # Hard constraint filtering & ranking algorithms
│ ├── fusion.js # EVM vector math (Valence, Arousal, Control)
│ ├── intentClassifier.js # Intent & safety guardrails
│ ├── personalization.js # User profiles & feedback loop learning
│ └── strategyEngine.js # Psychological strategy to nutrient mapping
├── data/
│ └── foods.json # Comprehensive food catalog with dietary tags & nutrients
├── pages/ # Next.js pages & API routes
│ ├── api/ # Endpoints (/api/suggestFood, /api/feedback, etc.)
│ ├── index.js # Main landing & recommendation dashboard
│ ├── about.js # System documentation & PWA installation guide
│ └── _app.js # Service worker registration & global providers
├── public/ # Static assets, icons, manifest.json, sw.js
├── tests/ # Native Node.js automated test suites
├── benchmark.js # Comprehensive offline accuracy & latency benchmark
├── LICENSE # Official MoodBite.AI Source-Available License
└── package.json # Dependencies and scripts
MoodBite.Ai is designed, engineered, and maintained by:
- Author: Ayush Singh
- GitHub: @SinghCod3r
- Repository: https://github.com/SinghCod3r/MoodBite.Ai
- Email: ayushsinghceee@gmail.com
Copyright (c) 2026 Ayush Singh. All rights reserved.
This project is licensed under the MoodBite.AI Source-Available License.
- Permitted: Personal and educational use, personal forks, educational modification, and research experimentation.
- Permission Required: Commercial use, incorporating into commercial products, public hosting/deployment of substantial services, redistribution, or licensing.
For commercial permissions or deployment inquiries, contact ayushsinghceee@gmail.com.
See the full LICENSE file for complete legal terms.