AxxessGuard

AI-Driven Preventive Health Partner

Built for the Axxess 2026 Hackathon

---

Overview

AxxessGuard is an iOS application that acts as a proactive, AI-powered personal health companion. It collects real-time vitals from Apple HealthKit, analyzes them using a large language model, detects potential falls, and sends smart health alerts; all in one unified experience!

---

Features

Feature	Description
📊 Live Vitals Dashboard	Real-time heart rate, step count, and SpO2 pulled from HealthKit
🧠 Predictive Risk Analysis	AI assesses your vitals and returns a LOW / MODERATE / HIGH risk score
💬 Virtual Health Assistant	Conversational AI chatbot for symptom triage, medication questions, and health guidance
🥗 Lifestyle Coach	Personalized AI-generated diet and exercise plans based on your vitals and health goals
🚨 Smart Notifications	Automatic alerts for high/low heart rate, hypertensive crisis, blood sugar anomalies, and daily summaries
🤸 Fall Detection	Accelerometer-based fall detection using a two-phase freefall + impact algorithm

---

Project Structure

AxxessGuard/
│
├── App
│   ├── AxxessGuardApp.swift          # App entry point (@main)
│   └── AxxessGuard.entitlements      # HealthKit entitlements
│
├── Config
│   └── Info.plist                    # App permissions, ATS config, background modes
│
├── Views
│   ├── ContentView.swift             # Root TabView + custom tab bar
│   ├── RiskModelView.swift           # Predictive risk AI analysis screen
│   ├── VirtualAssistantView.swift    # Chat-based health assistant screen
│   └── LifestyleCoachView.swift      # Diet & exercise plan generator screen
│
├── Managers
│   ├── HealthKitManager.swift        # HealthKit authorization, queries, live HR session
│   ├── NotificationManager.swift     # In-app notification logic + health alert triggers
│   └── FallDetectionManager.swift    # Accelerometer fall detection (freefall + impact phases)
│
├── Services
│   └── FeatherlessService.swift      # AI API client (REST, no external dependencies)
│
└── UI Components
    └── NotificationViews.swift       # Notification bell, panel, banner, and alert dialog views

---

Tech Stack & Frameworks

Apple Frameworks

Framework	Usage
SwiftUI	Entire UI layer — all views, animations, and layout
HealthKit	Reading heart rate, step count, SpO2; live workout session to keep HR sensor active
CoreMotion	Accelerometer access for fall detection at 50Hz sampling rate
UserNotifications	System push notification permission + scheduled daily 8AM summary
Combine	ObservableObject / @Published reactive data binding across managers
Foundation	Networking (URLSession), JSON parsing, date formatting

AI / Backend

Service	Usage
Featherless AI	Serverless LLM inference endpoint
Meta LLaMA 3.1 8B Instruct	The underlying model used for all AI features

All AI calls are made via plain URLSession REST requests — zero third party Swift packages required.

Architecture Pattern

• MVVM-adjacent: ObservableObject managers act as ViewModels; Views are purely declarative
• Async/Await: All AI and HealthKit network calls use Swift Concurrency (async/await, Task, MainActor)

---

Setup & Installation

Prerequisites

• Xcode 15+
• iOS 16+ device or simulator (HealthKit requires a real device for live heart rate)
• A Featherless AI API key → featherless.ai

Steps

1. Clone the repo

   git clone https://github.com/your-org/AxxessGuard.git
   cd AxxessGuard

2. Open in Xcode

   open AxxessGuard.xcodeproj

3. No package dependencies to install — the project uses no Swift Package Manager dependencies. Everything runs on native Apple frameworks + direct REST calls.

4. Add your API key Open FeatherlessService.swift and replace the API key:

   static let apiKey = "YOUR_FEATHERLESS_API_KEY"

5. Set your development team In Xcode → Target → Signing & Capabilities → set your Apple Developer Team (required for HealthKit on device)

6. Build & run on a real iPhone for full HealthKit + fall detection functionality

---

App Permissions Required

Permission	Reason
HealthKit — Heart Rate	Live and historical heart rate monitoring
HealthKit — Step Count	Daily activity tracking
HealthKit — Oxygen Saturation	SpO2 monitoring
HealthKit — Workout	Keeps the heart rate sensor active via a background workout session
Motion & Fitness	Accelerometer access for fall detection
Notifications	Health alerts and daily summaries

---

AI Features In Depth

Predictive Risk Model

Sends the user's vitals (HR, BP, glucose, steps, SpO2) to the LLM and asks it to return a 2–3 sentence risk summary ending with a structured RISK:LOW, RISK:MODERATE, or RISK:HIGH label, which the app parses to set the visual risk badge color.

Virtual Health Assistant

A multi-turn style chat interface. Each message prepends a system context block containing the user's current vitals before sending to the model. Responds in 2–4 sentences, always directing serious symptoms to a doctor. Never diagnoses.

Lifestyle Coach

Two separate AI calls. One for a diet plan, one for an exercise plan. Both are conditioned on the user's selected health goal (General Wellness, Weight Loss, Heart Health, etc.) and health condition (Diabetes, Hypertension, Post-Operative, etc.).

---

Fall Detection Algorithm

The FallDetectionManager uses a classic two phase detection approach on raw accelerometer data sampled at 50Hz:

1. Freefall Phase — Total acceleration magnitude drops below 0.35g (normal standing is ~1.0g)
2. Impact Phase — Within 600ms, magnitude spikes above 2.8g

Both phases must occur in sequence within the time window to trigger an alert. A 30 second cooldown prevents duplicate alerts from a single event.

---

Smart Notifications

Alerts are triggered automatically when vitals cross clinical thresholds:

Condition	Threshold	Alert Type
High Heart Rate	> 120 BPM	🔴 Critical
Elevated Heart Rate	100–120 BPM	🟠 Warning
Low Heart Rate	< 50 BPM	🟠 Warning
Hypertensive Crisis	Systolic > 180 or Diastolic > 120	🔴 Critical
Stage 2 Hypertension	Systolic ≥ 140 or Diastolic ≥ 90	🟠 Warning
Critical High Glucose	> 250 mg/dL	🔴 Critical
High Glucose	> 125 mg/dL	🟠 Warning
Low Glucose	< 70 mg/dL	🔴 Critical
Fall Detected	Algorithm trigger	🔴 Critical
Daily Summary	Step count based	🔵 Info / 🟢 Success

---

Hackathon Track Alignment

Requirement	Implementation
Wearable Integration	Apple Watch via HealthKit — live HKWorkoutSession keeps the Watch heart rate sensor continuously active; HKAnchoredObjectQuery streams real-time HR, step count, and SpO2 directly from the Watch to the app
Predictive Risk Modeling	LLaMA 3.1 risk assessment with structured output parsing (RISK:LOW / MODERATE / HIGH)
Virtual Assistant	Full chat UI with health context injection per message
Smart Alerts	NotificationManager with clinical thresholds + banner UI
Diet & Lifestyle Coaching	Goal/condition aware AI plan generation
Fall Detection (bonus)	Two phase CoreMotion accelerometer algorithm

---

Use Cases

These are directly supported through the condition selectors in the Lifestyle Coach and the AI's context-aware responses across all features:

Use Case	How AxxessGuard Addresses It
Chronic Disease Management	Users can select Diabetes or Hypertension as their health condition. The AI tailors diet plans, exercise recommendations, and risk assessments specifically around managing those conditions. Blood glucose and blood pressure are tracked as first-class vitals.
Elderly Care	"Elderly Care" is a selectable condition in the Lifestyle Coach, prompting the AI to recommend low impact, safe exercises and easy-to-prepare meals. Fall detection runs continuously in the background and fires a critical alert the moment a fall is detected, with an option to alert emergency contacts.
Post-Operative Recovery	"Post-Operative" is a selectable condition, causing the AI to generate conservative recovery appropriate plans, avoiding strenuous activity and recommending nutrition that supports healing. The risk model also factors this in when assessing vitals.

---

Future Features

• Add age, weight, and sex input so AI model has better context to work with
• Improve conversation history for the chatbot — pass previous messages to the AI so it remembers context within a session
• Fix some notification issues when entering user input
• Tweak fall detection algorithm
• Onboarding flow for profile input
• Historical vital charts shwoing range from past days
• Possible watch app companion

Devpost & Video Demo

https://devpost.com/software/axxessguard https://www.youtube.com/shorts/ttEiJdkgv7Y