Edge AI + Cloud ML Architecture
A two-tier AI infrastructure built for real-time on-device inference and population-scale wellness analytics.
The Physical Platform Behind the AI
The TCM-1 AI layer runs on manufactured hardware built by DMT, our manufacturing partner. Shown below: a product render of the assembled unit, and a bench demo of the working Edge AI Engine board.
End-to-End AI Pipeline
From the client's device to the practitioner's dashboard, encrypted in transit, built to an IEC 62304 software framework, and designed with HIPAA/GDPR/Malaysian PDPA alignment in mind. Full pipeline in development.
Two-Tier AI Infrastructure
Two-tier AI infrastructure designed for low-latency on-device inference and population-scale cloud analytics, built to an IEC 62304 software lifecycle in development.
Edge AI: On Device
Real-time AI inference on-device. No cloud dependency for session monitoring, targeting low-latency response for all detection models.
Cloud ML: Population
High-performance cloud infrastructure for population model training. Designed to aggregate anonymised session data to build outcome analytics at scale.
IEC 62304 Framework
All AI code is being developed under an IEC 62304 software lifecycle standard, with change control, risk traceability, and regulatory documentation built in from the start. In development, not yet independently audited.
Practitioner Dashboard
SaaS layer planned for practitioners: session analytics, adherence monitoring, and population-level wellness insights. In development.
AI Development & Validation Framework
Every AI model on the TCM-1 platform is developed under a documented, auditable process, not shipped as a black box. This framework is being built out alongside the models themselves; it is in development, not yet independently audited.
IEC 62304 Software Lifecycle
Requirements traceability, change control, and risk analysis will be applied to every model release, following the same software lifecycle standard our manufacturing partner uses for embedded device firmware.
IMDRF SaMD Risk Classification
Model risk is categorised against the International Medical Device Regulators Forum's Software as a Medical Device framework, the reference standard also guiding our manufacturing partner's own AI-enabled device roadmap.
Dataset Governance & Locked Models
Training data provenance is tracked per model version, and a model is version-locked with a documented validation record before it can be deployed, so what ships matches what was tested.
Four AI Models. TCM-1 Platform. (All In Development)
Edge AI: On Device
In Development
Model A: Session Detection
1D CNN + LSTM architecture for on-device Edge AI inference. Detects and classifies TCM session events in real time with a low-latency target. Foundation model for the TCM-1 platform. All other edge models build on it.
Edge AI: On Device
Planned
Model B: Session Consistency
Transformer-based anomaly detection. Flags unusual session patterns (missed sessions, off-protocol delivery) for practitioner review. Aims to help ensure repeatable, consistent TCM session delivery.
Cloud ML: Population Analytics
Planned
Model C: Personalised Sessions
Temporal Fusion Transformer for Cloud ML. Learns individual user session-response patterns from real-world data, generating personalised session recommendations updated periodically as more data is collected.
Cloud ML: Population Analytics
Planned
Model D: Population Analytics
Population-level session analytics model. Aims to quantify session adherence and self-reported wellness correlation across user cohorts, the proprietary data asset that will deepen as TCM-1 sessions are recorded.