health & longevity
The India data gap in longevity
Why Western-skewed health models are not enough, and what it would take to build longitudinal health data for India and South Asia.
The question behind Notch
The question behind Notch is not "can India build another wearable?"
That would be too small, and honestly not very interesting. The world has enough devices that count steps, show notifications, and produce charts that make healthy people feel slightly more informed than they were yesterday.
The more interesting question is this: what happens if the models that will guide preventive health for the next generation are trained mostly on people who are not us?
That question sounds abstract until you slow it down. A health model is not magic. It is a compression of what it has seen. If it has mostly seen Western bodies, Western lifestyles, Western food patterns, Western sleep environments, Western disease progression, Western clinical access, and Western longitudinal datasets, then it will carry those assumptions into its predictions. Some of them will generalize. Some will not. The dangerous part is that we may not know which is which until the model is already trusted.
India and South Asia are too large to be a validation set.
This is the claim I keep returning to.
Why population data matters
The body is universal in one sense and local in another. Everyone has a heart. Everyone sleeps. Everyone ages. But the way health unfolds is shaped by genetics, diet, climate, pollution, work, stress, family structure, medical access, and habits so ordinary that they rarely enter product requirements.
This is why data matters. Not because data is fashionable, but because health is full of baselines. A heart rate reading means little alone. Sleep duration means little alone. Temperature, recovery, movement, and blood markers all become more useful when you know what is normal for this person, this population, and this context.
Most consumer wearables avoid this problem by staying shallow. They show trends, badges, rings, and generic advice. That is useful up to a point. But the next step in health technology will not be better graphs. It will be interpretation. It will be models that notice patterns early, personalize recommendations, and eventually help clinicians and individuals reason about risk before disease becomes obvious.
That step requires longitudinal data.
And not just any longitudinal data. It requires data that represents the people who will depend on the output.
The India data gap is not a branding problem. It is an epistemic problem. If you do not measure a population, you will eventually pretend another population is close enough.
The boring beginning
The romantic version of this project would start with a foundational model. The practical version starts with a band.
That can sound disappointing. A band is small. It has sensors, firmware, a strap, a battery, a charging path, a mobile app, and many annoying ways to fail. But that is exactly why it is the right beginning. A dataset does not appear because someone writes "data moat" in a pitch deck. It appears because real people agree to wear something repeatedly, and the device quietly earns the right to keep collecting.
Hardware is not the final asset. Trust is.
The hardware is the instrument through which trust becomes data.
That sentence has become a useful filter for me. If a hardware decision improves specs but reduces trust, it may be a bad decision. If a feature makes the app impressive but makes the user anxious, it may reduce long-term data quality. If the device collects more signals but becomes uncomfortable to wear, the extra sensor is not an upgrade. It is a tax on continuity.
Continuity is the hidden requirement in health.
A one-time measurement can diagnose some things. But aging, recovery, metabolic change, stress load, cardiovascular drift, and sleep debt are stories. They need time. They need boring repetition. They need the user to keep participating after the novelty disappears.
What the big companies will miss
Google, Apple, Samsung, and the rest are not stupid. They have enormous advantages: distribution, silicon, design, operating systems, cloud infrastructure, and existing user trust. It would be naive to pretend a small Indian hardware project can outspend them.
But big companies often miss what is not legible at their scale.
They like global products. They like clean categories. They like features that can be explained in one launch slide. They like datasets that fit legal, commercial, and operational structures they already understand. India is harder than that. South Asia is not a single user persona. The interesting health questions here cut across language, income, diet, family behavior, urban pollution, rural access, heat, religious practice, and medical habits that do not map neatly onto California product thinking.
This does not mean global companies cannot serve India. It means the first truly useful health model for India may need to be built from closer contact with Indian lives.
The work is not merely to localize the interface. It is to localize the assumptions.
That is slower. It is also more defensible.
The CSDN lesson
One thing I noticed while reading Chinese technical blogs is that many good posts are not afraid to be practical. They do not treat implementation details as beneath the idea. A post about wearables will often move from background to architecture to data flow to scenarios to problems. It can feel mechanical, but the instinct is correct: technology becomes real through steps.
That is a useful correction to startup writing, which often floats too high.
So here is the practical chain for Notch as I currently see it:
- Build a wearable people will actually keep wearing.
- Start with a small set of reliable signals instead of pretending to measure everything.
- Create a data pipeline that preserves context, not just numbers.
- Learn what causes missing, noisy, or misleading data in real Indian use.
- Use the first cohort to improve hardware, onboarding, and interpretation.
- Expand signals only when the previous layer is trustworthy.
- Treat the dataset as a long-term scientific asset, not a side effect of selling devices.
None of these steps is glamorous. That is why they matter.
The hard part is not AI
The hardest part of building an India-focused longevity model is probably not the model.
The hard part is earning enough high-quality, long-term participation from enough people that a model becomes worth training. The hard part is consent, retention, sensor quality, missing data, device comfort, battery behavior, repair, privacy, clinical usefulness, and the discipline to avoid making claims before the data supports them.
This is where useful writing and useful building resemble each other. You have to make claims as strong as possible without making them false. The same standard should apply to health products. Say what you can measure. Say what you cannot. Do not hide uncertainty, but do not use uncertainty as an excuse to say nothing.
India does not need another wellness gadget that turns weak signals into confident advice.
It needs health infrastructure that begins humbly enough to become accurate.
That is why the data gap matters. Not because "India" is a convenient market story, but because a population that is not measured carefully will be understood lazily. And lazy understanding, when embedded into health systems, becomes harm with a clean interface.
The opportunity is large because the missing layer is basic.
We need to see ourselves clearly enough to build from there.