Multi-modal language
Bhashini-aligned regional language workflows for Assamese, Manipuri, Bodo, and Mizo, paired with quantized speech recognition for offline-first intent capture.
Technical architecture
Intelligence that stays useful offline.
MEDICHARM is designed for remote North Eastern territories where care cannot pause when bandwidth does. Local capabilities take priority; cloud sync improves continuity when trusted connectivity returns.
Local-first intent engine
A quantized Llama 3 8B or Mistral 7B-class model can reside on-device or on a local gateway for low-latency caregiver assistance.
Resilient edge care
Core sessions, reminders, and safety workflows remain useful during intermittent or absent connectivity.
Edge–cloud split
What happens locally, and what waits for sync
| Feature | Local processing · offline | Cloud synchronization · online |
|---|---|---|
| Data storage | Encrypted local data store for daily care logs | Secure caregiver backup and reviewed sync |
| Voice interaction | On-device reminder prompts and offline intent capture | Language updates and consented model improvements |
| Safety alerts | Local panic flow and care plan access | Verified push or SMS routing to remote kin |
| Cognitive games | Core logic and low-bandwidth activity assets | Performance analytics and care-team reporting |
Natural-language support
Intent without cloud latency
“Where is my daughter?”
Reassurance and familiar-contact prompt
“Play my favorite Borgeet music.”
Reminiscence session cue
“What medicine do I take now?”
Visual and audio medication prompt
The recall activity uses the provisioned voice capability to play its prompt aloud. The architecture remains compatible with a real-time voice agent for more conversational care support.
Trust by design
Protect care data at every layer
At rest
AES-256 protection for locally stored sensitive data.
In transit
TLS 1.3 and certificate pinning for secure edge–cloud communication.
HIPAA-aligned safeguards
Minimum-necessary data sharing, consent-led caregiver access, role-based access controls, and audit-ready access logging are the production privacy design baseline; covered deployments also require the appropriate Business Associate Agreements.
India health-data alignment
DISHA-informed privacy practice, explicit consent, and data-minimization principles guide regional deployment.
Sovereignty
Local-first processing minimizes unnecessary transmission of health context.