A phone-native stack. AI for judgment. Physical AI for perception.
Chakshus is engineered as a smartphone application first. Every layer above — AI, Physical AI, integration, security — exists to make the experience in the user's hand fast, reliable, and defensible.
Smartphone Application Layer
The Chakshus app ships natively on iOS and Android, distributed through the App Store, Google Play, MDM, or enterprise app catalogs. The UX is built for one-handed operation, glove-friendly tap targets, and screen-on-time discipline. State is local-first: users continue working in tunnels, basements, and remote sites; deltas reconcile on reconnect with conflict-aware sync. Push notifications are role- and shift-aware. Biometric unlock, app-level PIN, and remote wipe are standard. Role-based flows mean a technician, supervisor, and plant manager see different surfaces of the same install.
AI Layer Inside the App
AI runs in two places: on the device for low-latency, privacy-sensitive tasks (text classification, intent parsing, summarization, redaction) and in the Chakshus cloud for heavier reasoning (planning, multi-document synthesis, recommendations). The app routes intelligently based on connectivity, payload sensitivity, and battery state. Models are versioned per tenant, with an evaluation harness that gates promotions. Outputs are explainable: every suggestion ships with the source it leaned on. No customer data trains shared models without explicit opt-in.
Physical AI Layer Inside the App
Physical AI uses the phone's camera, microphone, IMU, and where available LiDAR to perceive the physical world in real time. On-device vision models detect parts, read gauges, recognize tags, and segment scenes at interactive frame rates. Audio models pick up bearing whine, leaks, and alarm tones. IMU-derived motion features classify operator posture and equipment vibration. Outputs are anchored to the user's view via on-screen overlays, so guidance lives where the work is. Confidence thresholds are tunable per site.
Generic categories. No vendor logo walls.
Chakshus connects to enterprise categories rather than naming specific vendors. Connectors are configured per tenant, with field-level mapping and transformation.
Phones primary. Tablets secondary.
- Primary: iOS and Android smartphones — current OS and one prior major version.
- Secondary: iOS and Android tablets, supported for supervisor and review surfaces.
- Camera, microphone, IMU required for Physical AI features; LiDAR optional and used when present.
Privacy by design.
- Tenant isolation by default; per-tenant encryption keys.
- TLS 1.3 in transit, AES-256 at rest, secure enclave for credentials.
- SSO, SCIM, RBAC, biometric unlock, remote wipe.
- On-device processing for sensitive payloads, configurable redaction.
- Alignment with [STANDARD: e.g., SOC 2, ISO 27001] [Status: confirm before publishing].