MoodFood

Converts user-reported emotional states into personalized food suggestions through a conversational chatbot interface that captures mood context, intensity, and triggers. The system likely uses a multi-step inference pipeline: mood classification (happy, stressed, anxious, tired, etc.) → contextual enrichment (time of day, recent activities, dietary restrictions) → recommendation ranking via a mood-food correlation model trained on user behavior patterns and nutritional science heuristics. The chatbot maintains conversational context across turns to refine recommendations without requiring explicit structured input.

Implements a natural-language chatbot that guides users through mood capture without requiring explicit form submission. The chatbot likely uses intent recognition (via NLU or LLM-based classification) to extract mood keywords, intensity, context, and triggers from free-form text input. It maintains conversation state across multiple turns, asking clarifying follow-up questions (e.g., 'Is this stress from work or personal life?') to enrich the mood profile before generating recommendations. The interface abstracts away structured data entry, making mood logging feel like a casual conversation rather than a clinical assessment.

Builds a user-specific model of mood-to-food associations by aggregating historical mood logs and food recommendations over time. The system likely tracks which food recommendations users accept/reject, paired with their reported mood state, to learn individual preferences (e.g., 'User tends to prefer comfort foods when stressed, but lighter foods when anxious'). This personalization layer may use collaborative filtering (comparing user patterns to similar users) or content-based filtering (matching mood-food pairs to nutritional/sensory properties). The model improves recommendation relevance as more data is logged, but requires sufficient historical data (cold-start problem) to become effective.

Filters food recommendations based on user-reported dietary restrictions, allergies, and preferences while maintaining mood-relevance. The system likely maintains a constraint satisfaction layer that intersects mood-based recommendations with a user's dietary profile (vegetarian, gluten-free, nut allergy, calorie limits, etc.). This prevents recommending foods that match the mood but violate dietary constraints. The filtering may also consider time-of-day context (breakfast vs. dinner recommendations differ) and meal type (snack vs. full meal) to ensure recommendations are contextually appropriate.

Maintains a persistent log of user mood entries and food recommendations over time, enabling historical analysis and trend detection. The system stores mood state, timestamp, context, recommended foods, and user acceptance/rejection signals. It then generates insights by analyzing patterns: identifying recurring mood-food associations ('You eat pasta when stressed'), detecting seasonal or temporal trends ('Your stress levels spike on Mondays'), and surfacing behavioral patterns ('You reject salads when anxious, but accept them when happy'). Insights are likely presented as natural-language summaries or visualizations (charts, heatmaps) to help users understand their emotional eating habits.

Implements a freemium business model where core mood-logging and basic recommendations are free, with premium features (advanced insights, export, priority support) behind a paywall. The system likely gates features at the API or UI level, checking user subscription status before allowing access to premium endpoints. Free users may have rate limits (e.g., 5 mood logs per week) or feature restrictions (e.g., insights only available to premium users). This model reduces friction for user acquisition while monetizing engaged users who derive value from the service.