Body Mode - Bio-Adaptive AI Health Companion

Inspiration

Most health apps give you a plan and blame you when real life breaks it.

Body Mode does the opposite.

Powered by Google Gemini 3 Flash, Body Mode is a phone-first adaptive health agent that builds your day from live signals, routine, goals, meals, sleep, work patterns, family context, health constraints, habits, and real-world environment,. Then it keeps rewriting that day as life changes.

It works in phone-only mode using on-device sensors. Optional wearable data through Health Connect makes it even stronger. The phone is the base. Wearables are the amplifier.

We asked one question deeper than "what if an app tracked your health?"

What if a system actually understood your life?

Not just your steps. Not just your calories. Your recovery. Your family. Your culture. Your goals. Your body on its worst day and your most important week. A system that does not fight reality — it reads it and responds.

That is what we built.

Important for judges: The Android app is the full product experience. The web version is a lightweight demo. Body Mode works in phone-only mode using on-device signals, and optional Health Connect / wearable data further improves personalization when available.

APK: https://bodymode.netlify.app/body-mode.apk

What it does

Body Mode is a bio-adaptive AI health agent that builds your day, watches reality break it, and rebuilds it in real time — powered entirely by Google Gemini 3 Flash.

The Four Pillars

1 — Phone-first intelligence

Body Mode works in phone-only mode using motion, activity recognition, location patterns, WiFi context, timing behavior, and sensor-based sleep/wake inference. No smartwatch required. When Health Connect wearable data is available, the system becomes even richer — but never depends on it.

2 — 90-day adaptive memory

Body Mode carries up to 90 days of evolving context. The app and Gemini 3 actively compress the past, strip repetition, discard dead patterns, and keep only what still improves the next plan. Memory stays sharp. Plans keep improving.

Over time the system learns what you actually eat, when your sleep breaks, how your workday behaves, which habits lift you, which quietly drain you, and what your intake patterns appear to be lacking over time. Then it uses that understanding to surface nutrition gaps, recovery pressure, and routine breakdowns — and responds with better plans grounded in your country, your food culture, and the kind of life you are actually living.

3 — Plans built for real human lives

Body Mode does not generate ideal plans for perfect days. It generates followable plans for real human lives.

Sometimes the right move is not "go train harder." Sometimes it is: take your kids to the park, walk with them, lower stress, move your body, and still hit the recovery goal.

The system understands your family context, your time constraints, your cultural reality, and what kind of week you are actually in:

Low-state or recovery periods — When the user is unwell or in a low-energy phase, the plan shifts toward recovery: lighter movement, supportive nutrition, adjusted hydration, reduced intensity.
Cycle-aware planning support — Energy, cravings, and recovery capacity shift across hormonal phases. The plan adapts with them.
Performance or competition week — Peak output timing, carbohydrate strategy, pre- and post-event recovery nutrition, sleep prioritization.
Deep work or exam season — Cognitive nutrition moves forward. Stable blood glucose, focus-supporting meals, and movement scheduled to enhance concentration rather than drain it.
Appearance or wellness goals — Nutritional lens shifts toward skin-supporting, anti-inflammatory, and micronutrient-dense food choices over time.

The goal changes. The entire system recalibrates.

4 — Plan → Reality → Replan

$$\text{Plan} \rightarrow \text{Reality} \rightarrow \text{Replan}$$

Unplanned meal. Missed workout. Bad night. Cycle shift. Low-state day. Schedule change. Body Mode reacts immediately. This is the core loop. This is what turns a static plan into a living system.

Full Feature Set

Daily Plan Generation — Health context, goals, routine, work schedule, sleep pattern, family reality, dietary preferences, and live bio-state. Not a template. A day built around your actual life.
Real-Time Replanning — Any deviation triggers immediate Gemini 3 recalculation of the remaining day.
Nutrition Gap Tracking — Multi-day intake patterns tracked. Gaps trigger automatic meal and food adjustments.
Cultural Food Intelligence — Recommendations fit the user's country, food culture, and realistic living environment.
Multimodal Food Analysis — Camera-based meal scanning estimates calories, macros, and 30+ micronutrients from a photo or short video.
Smart Fridge Scanner — Visible ingredients photographed. Gemini generates recipes that fill today's nutrition gaps.
AI Coach with Shared Context — The coach sees the same plan, intake, and bio-state as the planner. Context-aware guidance, not generic advice.
Sleep and Wake Intelligence — Phone sensors infer sleep and wake transitions. Gemini adjusts recovery logic and next-day structure accordingly.

Bio-Context Pipeline

BioEngine converts raw signals into three guardrails injected into every Gemini 3 prompt:

$$\text{Neural Battery} \quad \text{Hormonal Load} \quad \text{Physical Fatigue}$$

Instead of: "Here is the ideal schedule." Body Mode says: "Here is the best possible day for the body you actually have right now — and the goal you are actually working toward."

5-Sensor Context Fusion — Accelerometer + GPS + WiFi learning + Activity Recognition + Magnetometer detect whether the user is sleeping, commuting, at the gym, working, or shifting context in meaningful ways.

Android System Integration

Floating Overlay Reminders — Appear over any app. The plan escapes the interface and reaches real life.
AlarmManager Scheduling — Meals, hydration, movement, and recovery timing survive background and foreground.
Auto Sleep / Wake Inference — Phone sensors support ongoing sleep/wake inference.
Boot Recovery — Plans and schedules survive device restart.

4-Tier Adaptive System

Tier	Mode	Description
1	Full AI	Gemini 3 Flash with full live context injection
2	Reduced AI	Fewer Gemini calls under constrained conditions
3	Rule-Based	Offline heuristics when network drops
4	Manual	User-driven fallback

Body Mode remains usable across degraded conditions instead of collapsing when reality becomes imperfect.

How we built it

Phase 1 — Google AI Studio We validated whether Gemini 3 Flash could act as the central reasoning layer for structured planning, multimodal food understanding, context-aware coaching, goal-specific adaptation, and adaptive replanning — before writing a single line of mobile code.

Phase 2 — Local Development

React Native + Expo for the mobile layer
TypeScript across all product logic
Netlify Functions for secure Gemini API proxying
Room DB via native bridge for offline storage
Health Connect SDK for optional wearable augmentation

Phase 3 — Android End-to-End

Native bridges: SleepBridge, OverlayBridge, LocationBridge, TxStore
Gemini fallback chain: gemini-3-flash-preview → gemini-flash-latest → gemini-flash-lite
Circuit breaker + exponential backoff for API resilience
LLM Queue Service with priority and energy gating across multiple job types
13-language internationalization with culturally-aware meal suggestions

Challenges we ran into

Sensor inconsistency — Real-world signals are messy and delayed. Freshness-aware blending keeps the system stable without depending on any single source.
Sleep vs inactivity ambiguity — Distinguishing true sleep from stillness required multi-signal confidence scoring.
Goal-state transitions — Shifting between competition week, exam season, and recovery mode required clean nutritional priority reweighting without losing personal context.
Prompt constraint design — Raw user context had to become wellness-aware decision inputs before reaching Gemini 3.
Offline resilience — Four degradation tiers keep the system usable when connectivity drops.
Android background survival — Overlays, reminders, and sleep-aware logic had to survive outside the foreground app.

Accomplishments that we're proud of

A true adaptive loop — plan, reality, replan, continuously
90-day compressed memory — the system learns and stays sharp over time
Phone-first intelligence — no smartwatch required
Goal-aware recalibration — tournament, exams, recovery, wellness goals, cycle phases — the entire plan shifts when the user's focus shifts
Cultural food realism — recommendations fit where you actually live
Family-aware planning — goals combined into moments that fit real life
One reasoning engine powering the entire adaptive loop — Gemini 3 Flash end-to-end

What we learned

The real challenge in health AI is not generating advice — it is keeping advice alive when life gets messy
Phone sensors alone can go surprisingly far
Wearables make the system richer, not merely possible
The difference between a tracker and a companion is adaptation
The difference between a plan and a living system is whether it survives reality — illness, cycle shifts, bad days, shifting goals, and all