What the architecture supports

GaitAI processes human movement, so privacy and access control are part of the pipeline rather than a layer on top of it. This page documents what the architecture supports, control by control, in the language a security or procurement reviewer needs: what is available, what is configurable, and what remains the deploying organisation's decision.

Evidence, privacy, security and responsible deployment are collected together in the Trust Center.

Control areas

Processing location

Designed for edge and on-device inference as well as server-side processing; the optimised edge pipeline is the subject of the granted patent. Which mode a given site runs is a deployment decision.Documented in: Patent 402202 · Edge inference capability

Identity-not-required workflow

The architecture is designed to support task-relevant pose and movement features in workflows that do not require identity matching. Pose, gait and trajectories may still carry identifying information: privacy-aware architecture does not automatically mean anonymity.Documented in: PrivacyGuard · pipeline stages

Face blur

Face blur is designed as an optional pipeline transformation stage that can be applied before analytics rather than after.Documented in: PrivacyGuard · processing pipeline

Raw video handling

The pipeline is designed to transform raw streams into movement features, so analytics operate on those features rather than on raw footage.Documented in: PrivacyGuard · privacy stack

Retention controls

Retention is intended to be configurable per data class, with schedules set per site, so clinical captures need be kept only as long as the care workflow requires. Actual periods are agreed per deployment.Documented in: PrivacyGuard · deployment · shared clinical privacy note

Role-based access

An authorization layer is designed to mediate dashboard views so roles see only permitted data, with policy configured per site and per role.Documented in: PrivacyGuard · access model

Audit logs

The design provides for logging of access and policy-change events, with audit logs exportable for an organisation's own compliance review.Documented in: PrivacyGuard · audit and policy logs

Consent

Clinical assessments are intended to be captured with informed consent, with policy and consent records held alongside the audit log.Documented in: Shared clinical privacy note · PrivacyGuard output schema

Data in transit

Captures are intended to be uploaded over encrypted channels. Transport and storage configuration is agreed per deployment.Documented in: Shared clinical privacy note

Biometric and identity processing

Identity-bearing modules (ReID, AccessMotion, Watchlist) are gated behind stricter policy controls and are intended only for deployments with lawful authority, consent, access control and auditability. Where identity-not-required movement workflows is sufficient, that is the intended default.Documented in: Responsible-deployment policy · Watchlist product record

Defence and other high-assurance uses

DefenceMotion is a SecureVision product module, configurable in Army, Navy and Air Force modes; Defence & Armed Forces is the documented deployment environment, which combines it with AccessMotion, PrivacyGuard and SuspiciousMotion for governed, human-supervised facility safety and authorised access. GaitAI is not designed for autonomous targeting or lethal decision-making; an access signal supplements existing credentials and never identifies or denies on its own. Modes are configurations of one module, not separate products. No defence deployment, customer or clearance is documented.Documented in: DefenceMotion product record · Defence & Armed Forces environment record · Responsible-deployment policy

Training data

Privacy-preserving handling of gait datasets inside deep-learning pipelines is a published research area (IET Biometrics, 2022).Documented in: Publication: gait dataset privacy protection

Scope of the controls

These are architectural capabilities of the GaitAI pipeline rather than a description of a running deployment; what a given site enables is agreed with it. Lawful basis, consent management and deployment decisions remain with the deploying organisation, and privacy-aware architecture is not a guarantee of anonymity.Documented in: PrivacyGuard · documented limitations

This website

Separate from the product controls above, here is what the site itself runs on — verified against the code rather than described in general terms.

Static delivery

Pages are pre-rendered and delivered over HTTPS by GitHub Pages. Separate services handle forms, comments, subscriptions and optional hosted Ask GaitAI answers; hosting providers may keep security logs.

Aggregate measurement

Article views and likes, and assistant interaction counts by page type, are stored in Firebase. The counter records contain no questions, uploaded media or visitor identifier. No advertising pixel, tag manager or session-recording script is present.

Form submissions

Demo and contact requests are delivered by Formspree, a third-party form processor. Fields are interest, name, work email, organization and message — no health or clinical data is requested.

Comments

New public comments store a display name, message, reply reference, time and moderation state in Google Firebase (Firestore), without collecting an email or sign-in identifier. Administrators can hide or delete comments. Cloudflare Turnstile may be loaded when configured; the browser challenge is not a server-enforced abuse guarantee.

Administration

Administrative data and moderation actions require an authorized, verified Firebase Authentication account. Opening the public sign-in page does not grant administrative access.

Browser storage

Browser storage holds preferences, like and duplicate-comment markers, a short assistant conversation and counter-deduplication markers. The Privacy Policy details which information is sent to service providers.

Ask GaitAI

Hosted answers send text questions, limited conversation context, page context and public record IDs to Cloudflare Workers AI. Application logs omit question and answer text; a daily IP-derived hash is used for rate limiting. The movement analyser does not send its video or pose data to this service.

What GaitAI does not claim

Stated explicitly so nothing above is read as more than it is. GaitAI does not currently hold or assert:

• SOC 2, ISO 27001 or any other security certification
• HIPAA, GDPR or DPDP Act compliance certification
• A specific guaranteed retention or deletion window
• Guaranteed encryption of data at rest
• Clinical approval, regulatory clearance or medical-device status
• Measured accuracy, latency or performance figures
• Replacement for clinician judgement or a standalone diagnosis
• Proof of identity from a movement match alone
• Guaranteed detection or prevention of safety incidents

Where a control needs to meet a specific regulatory obligation in your jurisdiction, that is a deployment conversation — the lawful basis, consent management and data-protection assessment stay with the deploying organisation.

Related reading

• PrivacyGuard — where these controls are applied in the SecureVision pipeline.
• PrivacyGuard technical view — pipeline stages, output schema and documented limitations.
• Responsible AI — how identity-bearing capabilities are governed.
• Deploy GaitAI — inputs, outputs, integration and pilot shape.

Enterprise deployments receive deployment-specific documentation. For details contact contact@gaitai.in. The contact form is the other way through.