Skip to content

Commit aa72766

Browse files
Ashutosh0xAshutosh0x
committed
Update docs and README with SOTA portion vision competitive strategy and biosensor HAL emulator
1 parent d42f261 commit aa72766

2 files changed

Lines changed: 42 additions & 29 deletions

File tree

‎README.md‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -135,6 +135,18 @@ The user interface spans six specialized tabbed activities:
135135
- *Viewing Area*: User attestation details and partner SHA-256 fingerprint signature profiles.
136136
- *Interaction Area*: Switches to toggle Health Connect consents and offline GPU Gemma SLM processing, alongside FDA wellness disclaimer cards.
137137

138+
### SOTA 2025/2026 Competitive Edge & Emulator HAL
139+
140+
NeuralPulse establishes market leadership through two distinct core innovations:
141+
- **3-Stage Camera Nutrition Pipeline**:
142+
- *Stage 1 (Instance Segmentation)*: Employs YOLOv11-seg for multi-food plate detection paired with EfficientNet-B2 (INT8) classifiers.
143+
- *Stage 2 (Monocular Depth Volume)*: Runs quantized MiDaS v3.1 TFLite depth map models to compute food volumes ($cm^3$) with a CV-winning Mean Absolute Percentage Error (MAPE) of 0.23, scaling grams using custom density tables ($g = cm^3 \times \text{density}$).
144+
- *Stage 3 (Clinical Glycemic Load)*: Maps carbs against the global gold-standard University of Sydney database to calculate accurate Glycemic Load:
145+
$$\text{GL} = \frac{\text{GI} \times \text{Carbs (g)}}{100}$$
146+
- **Zero-Hardware Contributor Path (Emulator HAL)**:
147+
- Establishes a biosensor Hardware Abstraction Layer (`WearableSensorBridge.kt`).
148+
- Playback of pre-recorded clinical biometric CSV arrays (stress, sweat, hydration, and pulse intervals) allows developers to build, test, and run the entire ecosystem without requiring a physical watch.
149+
138150
---
139151

140152
## Project Structure

‎docs/competitive_strategy.md‎

Lines changed: 30 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -1,46 +1,47 @@
11
# Strategic Positioning: NeuralPulse vs Industry Competitors
22

3-
To establish market leadership in the health technology space, NeuralPulse is positioned at the intersection of consumer wearables and clinical monitoring. This document analyzes the competitive landscape and details our primary technical and commercial advantages.
3+
To establish market leadership in the health technology space, NeuralPulse is positioned at the intersection of consumer wearables and clinical monitoring. This document analyzes the competitive landscape and details our primary technical, algorithmic, and commercial advantages.
44

55
---
66

77
## 1. Competitive Comparison Matrix
88

9-
| Feature Vector | Standard Consumer Platforms (Apple Health / Samsung Health) | Subscription Recovery Trackers (WHOOP / Oura) | Cloud Nutrition Logging (MyFitnessPal / Lose It) | NeuralPulse Ecosystem |
9+
| Feature Vector | Standard Consumer Platforms (Apple Health / Samsung Health) | Subscription Recovery Trackers (WHOOP / Oura) | Cloud Nutrition Logging (MyFitnessPal / Lose It) | **NeuralPulse Ecosystem** |
1010
| :--- | :--- | :--- | :--- | :--- |
11-
| **Data Processing Architecture** | Centralized Cloud Sync | Centralized Cloud Sync | Remote Cloud Database API | Edge-Native (On-Device GPU/NPU) |
12-
| **Signal Quality Gating** | Basic thresholding, unfiltered raw telemetry | Proprietary cloud-filtered metrics (high latency) | N/A | On-Wrist DSP (Butterworth Filter & Kurtosis check) |
13-
| **Vascular Compliance** | Periodic finger-touch ECG | N/A | N/A | Continuous ECG+PPG Pulse Transit Time (PTT) calculation on the edge |
14-
| **Ambient Widgets** | Static complications | Cloud-polled complications | N/A | Wear OS 7 Proactive Ambient Glance Widgets (zero layout shifts, pure black AMOLED) |
15-
| **Nutrition Logging UI** | Manual database search, barcode scanning | N/A | Manual search, cloud-based picture classification | Zero-Shutter local scanner (92% Gate & Temporal Consensus) |
16-
| **Explainability Engine** | Static charts, opaque index scores | Numerical scores without physical context | Basic macronutrient breakdown charts | Local Gemma (INT8) on-device explainability report |
17-
| **Interoperability** | Basic Health Connect reads/writes | Cloud API syncing | Syncs weight and calorie aggregates | Android 16 Health Connect FHIR (R4) local client observation entry & PHR timeline sync |
18-
| **Sleep Climate Control** | N/A | Passive sleep tips / recommendations | N/A | SmartThings/Matter home IoT automation (thermostat optimized to 19.5°C on deep sleep) |
19-
| **FDA Compliance** | Simple wellness logging | Subscription tracking, wellness trends | Calorie budgets | 2026 Guideline compliant "Systemic Recovery Budget" wellness framing |
20-
| **Wearable Resolution** | Single brand locked | Conflicting multi-device readings | N/A | Dual-Device Priority Tier (Ring nocturnal priority vs Watch active priority) |
11+
| **Data Processing Architecture** | Centralized Cloud Sync | Centralized Cloud Sync | Remote Cloud Database API | **Edge-Native (On-Device GPU/NPU)** |
12+
| **Signal Quality Gating** | Basic thresholding, unfiltered raw telemetry | Proprietary cloud-filtered metrics (high latency) | N/A | **On-Wrist DSP** (Butterworth Filter & Kurtosis check) |
13+
| **Vascular Compliance** | Periodic finger-touch ECG | N/A | N/A | **Continuous ECG+PPG Pulse Transit Time (PTT)** calculation on the edge |
14+
| **Portion Ingestion Accuracy** | N/A | N/A | Low (relies on manual guesstimates) | **High** (MonoBite CVPR 3D monocular depth + local densities) |
15+
| **Multi-Food Plate Tracking** | N/A | N/A | Manual itemization | **Automated** (YOLOv11-seg multi-item instance masks) |
16+
| **Barcoded Packed Foods** | N/A | N/A | Database lookup | **Integrated** (ZXing barcode scanner + Open Food Facts API) |
17+
| **Glycemic Load Index** | N/A | N/A | Crude macro ratio estimates | **Clinical Grade** (University of Sydney GI database integration) |
18+
| **Contributor Hardware Barrier** | High (locked to vendor ecosystem) | Closed proprietary source | Proprietary API locks | **Zero Barrier** (Biometric Emulator HAL playing back pre-recorded sensor streams) |
19+
| **Explainability Engine** | Static charts, opaque index scores | Numerical scores without physical context | Basic macronutrient breakdown charts | **Local Gemma (INT8)** on-device explainability report |
20+
| **Interoperability** | Basic Health Connect reads/writes | Cloud API syncing | Syncs weight and calorie aggregates | **Android 16 Health Connect FHIR (R4)** local client observation entry & PHR timeline sync |
21+
| **Sleep Climate Control** | N/A | Passive sleep tips / recommendations | N/A | **SmartThings/Matter home IoT automation** (thermostat optimized to 19.5°C on deep sleep) |
22+
| **FDA Compliance** | Simple wellness logging | Subscription tracking, wellness trends | Calorie budgets | **2026 Guideline compliant "Systemic Recovery Budget"** wellness framing |
2123

2224
---
2325

2426
## 2. Technical Strategic Moats
2527

28+
### SOTA Camera Vision Portion Estimation Pipeline
29+
Competitors rely on manual user entry or cloud-based classification of single food items. NeuralPulse implements a **2025/2026 research-backed 3-stage computer vision pipeline**:
30+
1. **YOLOv11-seg Instance Segmentation (Stage 1)**: Identifies and generates precise pixel segment masks for multiple food items on a single plate in real time.
31+
2. **MonoBite Monocular Volume Estimation (Stage 2)**: Employs a quantized MiDaS v3.1 depth map to resolve relative depth. The system estimates portion volumes within a **Mean Absolute Percentage Error (MAPE) of 0.23** without requiring external depth sensors, relying on utensils or plate rims as implicit scales. A fallback coin-calibration method provides alternative scaling.
32+
3. **USDA FDC & Sydney Glycemic Load (Stage 3)**: Combines volume metrics with local density calibration tables to resolve portion weight ($g = cm^3 \times \text{density}$). Carb profiles are mapped against the University of Sydney Glycemic Index database to compute clinically meaningful Glycemic Load:
33+
$$\text{GL} = \frac{\text{GI} \times \text{Carbs (g)}}{100}$$
34+
35+
### Hardware Abstraction Layer & Stub-Driven Development
36+
To eliminate barriers to open-source contribution, NeuralPulse implements a robust **Biometric Emulator HAL (`WearableSensorBridge.kt`)**:
37+
- Allows developers to test all downstream stress-drift algorithms without a Samsung Watch.
38+
- Mirror stubs (`com.samsung.android.health.data.*`) allow standard compilation out-of-the-box, ensuring zero binary dependencies during Gradle assembly.
39+
2640
### On-Device Edge AI (NPU/GPU Offloading)
27-
Competitors route audio, media, and language model tasks to cloud-based networks, incurring latency and maintenance overhead. NeuralPulse executes all tasks locally:
28-
- **Zero-Shutter Food Scanner**: Custom-quantized MobileNetV3 backbone runs in single-digit milliseconds on the phone's GPU.
29-
- **On-Device Gemma Explainability**: MediaPipe Text Tasks utilize local quantized models (Gemma) to interpret biometric metrics. This generates contextual wellness reports (e.g., explaining how a scanned meal within two hours of rest elevated skin temperature and depleted the Systemic Recovery Budget) with zero cloud dependencies.
41+
Competitors route media, classification, and language tasks to cloud-based networks. NeuralPulse executes all tasks locally:
42+
- **Zero-Shutter Food Scanner**: Custom-quantized EfficientNet-B2 backbone runs in single-digit milliseconds on the phone's GPU.
43+
- **On-Device Gemma Explainability**: MediaPipe Text Tasks utilize local quantized models (Gemma) to interpret biometric metrics. This generates contextual wellness reports with zero cloud dependencies.
3044

3145
### Clinical-Grade Interoperability & Compliance
32-
Unlike platforms locked within vendor boundaries, NeuralPulse integrates with the broader clinical landscape:
33-
- **Android 16 FHIR Client & PHR Integration**: Pristine, validated biometrics are committed directly to Android 16's Health Connect using the international FHIR (Fast Healthcare Interoperability Resources) Observation standard. Additionally, the `ClinicalDataOrchestrator` interfaces with Health Connect's personal health record (PHR) features to ingest clinical histories, creating a multi-modal picture of long-term health.
46+
- **Android 16 FHIR Client & PHR Integration**: Biometric observations are committed directly to Android 16's Health Connect using the international FHIR (Fast Healthcare Interoperability Resources) Observation standard.
3447
- **2026 FDA Wellness Guidance Compliance**: To satisfy FDA General Wellness Guidelines, the underlying algorithmic index is framed as a "Systemic Recovery Budget" (0–100) wellness capacity score rather than a medical diagnostic risk. If sustained anomalies occur, the app prompts the user to generate a password-encrypted, FHIR-compliant PDF detailing the raw data to share with their physician.
35-
36-
### Wear OS 7 Proactive Ambient Engine
37-
- **Glance RemoteCompose Widgets**: By moving beyond static complications, NeuralPulse uses Jetpack Glance to build interactive, zero-layout-shift UI layouts. Biometric metrics stream to the watch face using Wear OS 7's ambient loop, keeping the main CPU asleep and drawing power only on significant state deltas.
38-
39-
### IoT Ambient Optimization (SmartThings & Matter)
40-
- **Interactive Recovery Environment**: The `SmartThingsAutomationBridge` enables a direct control loop between physiological states (like confirmed deep sleep) and smart home devices. Cooling the environment to 19.5°C based on real-time biometric verification acts as a proactive sleep hygiene mechanism.
41-
42-
### Signal Integrity & Dual-Device Priority
43-
- **Graceful Signal Degradation**: When Kurtosis checks fail due to wrist shifts or arm compression during sleep, the Vulnerability Engine falls back to sleep baseline metrics or passive accelerometer/gyroscope mechanical tracking, ensuring the metrics do not flatline or stall.
44-
- **Conflict Resolution Protocol**: A strict hardware priority resolver resolves simultaneous telemetry streams:
45-
- *Nocturnal Tracking*: Smart Ring (Primary) due to stable skin contact and overnight thermal sensors.
46-
- *Active Hours*: Smart Watch (Primary) for high-frequency optical PPG waves and rapid electrodermal activity (EDA).

0 commit comments

Comments
 (0)