Heartbeat

Health apps provide reactive data — logs, charts, averages — but lack proactive intelligence.

• No app predicts that a baby will reject the 7pm bottle based on 3 weeks of feeding patterns
• No app detects that a senior's blood pressure trend will cross a threshold in 48 hours
• The missing layer is an AI that thinks strategically across time, not just responds to the current moment

Dual-brain architecture with polymorphic user profiles:

• Tactical LLM (Llama 3.1 8B via Groq, <500ms) — real-time decisions every interaction
• Strategic LLM (Llama 3.3 70B) — Deep Pulse analysis over 50+ accumulated heartbeats to detect patterns invisible at the individual level
• Polymorphic User DNA — single database schema with metaphor-driven UX: The Nest (baby), The Shield (celiac), The Library (student), The Greenhouse (senior)
• Same JSON columns carry completely different health semantics per profile type

Multiple interconnected engines working across the full health spectrum:

• Deep Pulse — detects multi-day patterns, produces typed insights (ALERTA_TENDENCIA, PREDICCIÓN, OPTIMIZACIÓN_RUTINA) with confidence scores
• Travel Engine — Haversine distance with hysteresis for travel state detection and flight inference (>1000km in <2h). Adjusts per profile: jet-lag mode for babies, gluten databases per country for celiacs, medication timezone shifts for seniors
• I2I Visual Evolution — image ancestry chain where each generation descends from the previous one with adaptive strength (0.1 subtle → 0.9 dramatic) reflecting real-time health state
• Crisis Simulation — multi-day stress scenarios exercising all three layers together
• Gamification — psychological reinforcement: points, streaks, level progression, and visual metaphors (growing gardens, fortified shields) for behavioral health loops

24+ API endpoints, 43+ LLM tool definitions, 4-provider AI abstraction with automatic fallback.