The problem#
Invip explored a simple but important idea: could a device translate visual context into spoken information for a visually impaired person?
Product direction#
The goal was not a generic assistant. It was a bridge between visual information and a user who could not rely on the screen — which is a much narrower and much harder product question.
A working prototype existed and was showcased. The work ran alongside her MS in Management of Technology at NYU Tandon, and the company was incorporated in the United States.
What she'd do differently today#
Modern multimodal models collapse pieces of that architecture into capabilities that are far easier to prototype. The product questions did not move: what information is useful, when should the system speak, how do you measure a wrong interpretation, and what happens when confidence is low?
Almost the same questions she cares about in applied AI now: capability is not enough; the product needs boundaries and reliability.
Frequently asked questions#
What was her role?
Co-founder and CTO. Invip was a U.S.-incorporated accessibility startup and an award-winning NYU project, built between 2015 and 2018 alongside her MS in Management of Technology at NYU Tandon.
Is Invip still running?
No. It is presented here as part of her timeline — a 2015–2018 venture — not as a live product. This portfolio does not claim active products it no longer operates.
Why keep an old accessibility project in the portfolio?
Because the hard questions transferred. Deciding what information is useful, when a system should speak, how to measure a wrong interpretation and what to do when confidence is low are exactly the questions an AI product has to answer. The capability got easier; the product judgment did not.
The same questions, a new stack
Capability is not enough; the product needs boundaries and reliability. That is the through-line from Invip to the current work.