Most of this problem is being solved backwards.

The name

वही — vahi — means "the same one."

It is an ordinary Hindi word. You would use it to point at someone across a room and say: that one. The same one. Not a similar one, not one who says they are — that one.

Which is the entire product in a single word, and the reason our mark is two points and the line between them. Two moments in a hiring process, and the proof that they are the same person. Remove the line and you have two unrelated dots, which is exactly what an unverified hire is.

The thesis

Detection decays. Binding does not.

Every deepfake detector is a model trained on the fakes that existed when it was built. Fakes improve; the detector's accuracy quietly erodes until someone retrains it. That is a treadmill, and whoever is on it has to keep running to stay in the same place.

Binding an identity is a different kind of claim. A government-anchored enrollment, a biometric template, a specific device, a specific session, a specific first day at work — these are facts that link to each other. A better fake does not un-link them.

So we treat detection models as replaceable parts. They plug into the product and get swapped as they age. The chain they plug into is the part we intend to still be right in ten years.

What we optimise for

The honest candidate should feel almost nothing.

Verification that punishes everyone to catch a few is bad product design and worse ethics. An honest candidate should encounter a link, a short check, and then get on with the interview they prepared for.

That shapes real decisions. Government identity checks happen once, not every round, because repeating them costs the candidate time and gains nothing. Candidates never pay — verification is a cost of hiring, not a toll on being hired. And the system cannot reject anyone; it can only hand a human evidence to look at.

Friction is a budget. We spend it where risk actually lives, and nowhere else.

India first

Built for the market with the problem and the rails.

India runs an enormous share of the world's remote technical hiring, has an organised proxy-interview economy operating in the open, and has government identity infrastructure that makes a real anchor possible. Those three facts do not overlap in many places.

Building for that environment first means designing for low-end devices, unreliable connections, and India's data-protection regime as the default rather than a later port. The product is portable — the identity rail and the policy rules are swappable per market — but the hard version of the problem is here, so we are solving it here.

Where we actually are

Pre-launch, and specific about it.

We would rather say this plainly than imply a maturity we have not reached. A company asking employers to trust its evidence about other people does not get to be vague about evidence concerning itself.