Platform

StarDeep, in plain language

StarDeep is NeuTigers' end-to-end platform for building health AI that runs on the device itself. It takes multimodal health data — wearable sensor streams, questionnaires, clinical records — and turns it into compact deep neural networks small and efficient enough to run inference on a smartwatch or other edge hardware, in real time, without sending personal data to the cloud.

The output is not a demo: it is a deployable model plus the surrounding data-quality, traceability and validation workflow needed to take it toward clinical use with partners.

Wearable-AI™

Wearable-AI™ is the name for what StarDeep produces: AI that runs locally on sensor-enabled devices — no cloud — for ultra-fast, private, real-time decisions.

• Private by design — HIPAA/GDPR compliant by design; no raw biometric data is transmitted off the device.
• No new hardware required — models deploy on smartwatches, rings, phones and embedded IoT devices people and health systems already use.
• Radically compact — Grow-and-Prune™ deep learning compresses models by roughly 1,000x without sacrificing accuracy.
• Edge-optimized performance — in internal benchmark results reported in our corporate materials, our edge-optimized algorithms outperform legacy neural networks with 1.2x the accuracy, 2x faster latency, 1,000x smaller model size and 1,000x greater energy efficiency.

From data to care: the StarDeep workflow

1. Data — acquisition of multimodal, real-world health data under consent, quality and traceability controls (GDPR-aligned).
2. Model development — training of compact, edge-optimized neural networks using NeuTigers' licensed model-efficiency and data-quality methods.
3. Validation — performance assessment on held-out and real-world data, and clinical studies with partner institutions where a program requires them; physiological and questionnaire data in our clinical work is collected under Institutional Review Board (IRB) approval.
4. Device deployment — packaging and optimization of the model for the target device, from smartwatches to other constrained hardware.
5. Care integration — delivering the model's outputs into partner products and care workflows, with monitoring and update paths.

Supported data and devices

StarDeep is designed to be device- and data-agnostic within the connected-health space. Work to date has centered on:

• Wearables and smartwatch sensors — commercially available devices capturing physiological signals such as those used in our SweetDeep™, CovidDeep and sickle-cell work.
• Multimodal health data — combinations of sensor streams, questionnaires and clinical data, which our methods are specifically built to fuse into a single compact model.

Why on-device matters

• Latency — insights are computed where the data is generated; no round-trip to a server.
• Privacy — raw physiological data can stay on the device, supporting privacy-preserving designs.
• Power — compact models built with our efficiency methods run within the tight energy budgets of wearables.
• Connectivity — programs can operate without continuous Wi-Fi or cellular coverage, which matters for decentralized studies and underserved settings.
• Scalability — deploying to devices people already own avoids per-user cloud compute costs and enables large populations to be reached through device partners.

What we do for partners

• Development — feasibility assessment, data strategy and custom model development on partner or jointly collected data.
• Validation — retrospective and prospective evaluation, including collaborative clinical studies with hospital partners.
• Deployment — optimization and integration of models onto partner devices and into partner applications and care pathways.

Princeton-origin technology core

StarDeep's differentiation rests on a portfolio of methods licensed from research originating at Princeton University:

• Grow-and-Prune — a training approach inspired by how biological neural networks develop, producing highly compact, accurate deep neural networks suited to low-power edge inference.
• CTRL — detects labeling errors in training data, improving the quality of the data models learn from.
• TUTOR — iteratively combines real and synthetic data to get the best-performing model from a given dataset — valuable in health, where labeled data is scarce.
• DOCTOR — enables multiple prediction "heads" on a single shared network core, so one compact model can address several conditions while conserving device memory and power.
• SCouT — an analytics framework for spatiotemporal health data aimed at more time- and subject-efficient clinical studies at comparable statistical power.

The same research lineage also includes cybersecurity frameworks (SHARKS and GRAVITAS) for securing constrained connected devices, and ongoing generative-AI work. These are not our commercial focus in health programs; they form part of the platform's defensibility and are available as a licensing layer for partners with adjacent needs.

Patents & IP

Key granted and filed patents protecting the platform:

• US Patent No. 17,613,284 — Grow-and-Prune: edge-AI learning architecture for patient-side, real-time biometric monitoring on wearables.
• International/US Patent PCT/US No. 19/41531 — SCANN: Synthesis of Compact and Accurate Neural Networks.
• US Patent No. 63,647,141 — DLN: Differentiable Logic Networks.
• US Patent No. 17,619,449 — DiabDeep: pervasive diabetes diagnosis based on wearable medical sensors and efficient neural networks.
• US Patent No. 18/235,422 — MHDeep: mental health detection based on wearable sensors and artificial neural networks.

These sit within a broader 2018–2027 IP portfolio spanning core edge-AI methods (NeST, ChamNet, H-LSTM, SCANN, DOCTOR, SCouT, DLNs), cybersecurity (SHARKS, GRAVITAS) and disease applications (DiabDeep, MHDeep, CovidDeep, SweetDeep, SickleDeep, CardioDeep, AZDeep), with additional applications filed.

One platform, a continuum of algorithms

The same wearable platform — a consumer smartwatch, the NeuTigers data-collection and management apps, and the StarDeep cloud — carries a growing family of disease algorithms. In metabolic health, SweetDeep™ screens for type 2 diabetes in the general population and MFDeep predicts fetal macrosomia risk in gestational diabetes. One device, one data pipeline, several clinically distinct programs — each with its own explicit evidence status.

Why NeuTigers is different

Four pillars of differentiation:

1. Proprietary edge-AI — patented algorithms and apps protect the technology, the data and the disease models.
2. Clinical knowledgebase — unique knowhow in decentralized clinical trials, large-scale sensor and clinical datasets, and a high bar for clinical proof.
3. On-device deployment — ultra-efficient on-device AI runs anywhere, with privacy by design.
4. Strategic ecosystem — a global wearable OEM, a top Ivy League university, and international clinical partners across the US, EU and MENA.

Technical diligence

We welcome technical and scientific diligence — methods, publications, study designs and deployment architecture. Start a partnership conversation or review the evidence library.