DekhbhaalPrivacy and safety
Your parent's words are theirs. The family sees how they are, never what they said, and nobody at Dekhbhaal reads them either. Here is how that works, and what happens on a hard day.
Who sees what
| Who | Sees | Never sees |
|---|---|---|
| The family, and brothers and sisters you invite | Talked today or no reply; medicines taken or not confirmed; a scam stopped and its kind; a hard moment; visits; stories your parent chose to send; Sunday's note | What your parent said, their memories, the messages they checked, what a paper said |
| People who come to the house | The medicines due on their visit, and visit times. Urgent alerts, if you allow it | Anything your parent said, scam checks, money, the weekly note, stories |
| A care company's desk | Urgent alerts, today's status, the numbers to call | Words, memories, medicines, visits, scam checks, the weekly note |
| Dekhbhaal's team | Counts, such as how many check-ins were answered this week | Anything anyone said or added |
Your parent decides
You set Dekhbhaal up and you may pay for it, but the data is your parent's. On their first day, in their own language, Dekhbhaal tells them what it is, what the family will see and what it won't, and what the people who come will see. Nothing is sent to them, and nothing is shown to you, until they say yes.
They are told on their screen each time someone new can see how they are, and they can remove anyone. Nothing they say reaches you unless they ask Dekhbhaal to send it and tap yes on the one message that goes.
How it's protected
- Encrypted with a key of its own. Everything your parent says, their voice notes, photos and memories, their medicines and appointments, the messages they check, and the names and numbers of the people around them are encrypted (AES-256) with a key that belongs to their account alone, before anything is stored. That key is itself protected by Google Cloud's key service.
- Stored in India. On Google Cloud in Mumbai.
- Never used to train AI. The AI runs on Google Cloud's Vertex AI, which doesn't train on customer data.
- No words in notifications or records. Alerts, notifications and our own system records carry ids and short status codes, never what anyone said.
- Delete means gone. When an account is deleted, everything stops at once. Three days later its key is destroyed, which makes all of it unreadable for good, and the data is deleted.
- No ads, no selling. Nothing is sold, and nothing is shown to advertisers.
The details are in the privacy policy.
On a hard day
Dekhbhaal reads what your parent says with the context of their life, so a joke stays a joke, and a story told for the third time is just a story. If it sounds like they may be in danger now, such as a fall, chest pain, sudden confusion, someone at the door, or a very low day, it:
- answers first with 112, word for word, and puts it on their screen as the biggest button, with your number and the neighbour's under it;
- tells them the family is being told, and what to do to stay safe until help comes;
- alerts you, and everyone with urgent alerts, at once. The alert says they may need help now, never why.
If the AI can't judge a message at that moment, a network error say, the message is never dropped. It is judged again within minutes, and a hard moment found late still alerts you.
Health questions
It reminds and it reads. It never advises: no diagnosis, no dose changes, no health tips, and no meaning for a test result. When your parent asks, it listens, and points them to their doctor and to you.
Always an AI
"AI hai, insaan nahi" sits under Dekhbhaal's name on every screen of your parent's phone. It never pretends to be a person, and never talks about a family, a home or a past of its own.