Hermet
Care that continues between visits.
Keeping care close.
Maria carries relevant context across calls, adapts to the pace of the conversation and prepares a summary when follow-up may be useful.
THE PRODUCT
Care teams and families cannot be present in every moment, and asking an older person to adopt a new app creates a barrier precisely where continuity matters. Hermet brings Maria, an AI companion, to the phone the person already knows how to use. There is no tablet, wearable or new interface to learn. On each call, Maria can return to relevant topics from earlier conversations, adapt to the person's pace and capture what they choose to share. When something may need attention, she prepares a follow-up summary. The family app and care team dashboard turn the conversation into visible, manageable context without presenting Maria as a replacement for professional care.
IN USE
- (01)
It starts on the phone
Maria calls the usual number. The person answers without installing anything and speaks at their own pace.
- (02)
Context carries forward
The next call retrieves relevant topics and allows anything misunderstood to be corrected or removed.
- (03)
Follow-up is prepared
If something may need attention, Hermet prepares a summary for the family and care team.
ENGINEERING
Maria runs as a real-time voice pipeline over standard telephony. A session orchestrator coordinates turn detection, voice processing, dialog generation and audio output with explicit latency, retry and duration limits. Before each call, the memory service assembles a scoped context window from the information permitted for that person. New memories follow specific write rules: not every utterance is retained, and any correction or deletion changes the context assembled for future calls.
The conversation manager treats silence, interruption, barge-in and partial transcripts as normal states rather than errors. A separate follow-up path turns relevant calls into structured summaries and surfaces uncertainty without issuing a diagnosis. Evaluations replay multi-call trajectories and measure context retention, correction propagation, turn latency, interruption recovery, alert precision and whether the summary contains enough evidence for a care team to decide the next step.
