[kartikkumar]

I definitely find myself having to wade through my photo and video collection just to find that one that I remember I took a few years ago but for the life of me can't remember exactly when and where (which is the easiest way to sort and find). I think the key for me would be how much training/how many questions it would take for the search to become accurate. It would also be important to me to be able to use this offline. Don't fancy having to upload all my personal photos and videos.

Have you done any user testing yet?

I can imagine that if you get the algorithm to work robustly that there might actually be an entirely different, enterprise market.

Do you have a proprietary algorithm? Or are you using something well-known, published/publically available?

Would this simply be a product that the end user pays for? Or do you have another revenue stream in mind?

[jeads]

I am very familiar with that process myself. :)

While a cached intermediate for offline use is a very real possibility, providing top quality service without use of our server system would not be feasible with the current device market. Server-side we can use hardware that is optimized for this kind of processing, without being constrained by device or browser limitations. Without that the service quality would degrade sufficiently to tarnish our product’s image and turn people away.

One example of this is the one you provided: the quantity of training required for accurate answers. If we need to keep this metric tight, relying on random ARM processors and sandboxed single-threaded javascript is not the way.

Datalba’s CEO is in the process of collecting early test users, and has been receiving large quantities of test media for use with the prototype from several of them.

Yes, there are many opportunities in the enterprise market for this. The immediate interest is in bringing a quality service to the world that many people will benefit from. The needs of enterprise customers are largely a subset of that goal, and the parts that aren’t can be handled in a given contract or agreement.

We have been working with and building on the latest in deep learning research, and have become quite proficient with designing and building these technologies.

While we will undoubtably be exploring payed options, I fully expect that to make this product a real success that there needs to be as few barriers to access as possible for the user. Degrading accuracy based on a paywall, the most seemingly economical choice, would alienate the majority of users before they have a chance become highly engaged.

Additional value-adds such as auxiliary services could be monetized, but so far we have managed to collect a number of expressions from consumers that this leads to a very ‘pushy’ presentation. While more market data is needed, it is clear this is territory that should be tread lightly.

The remaining option that is most clear is to monetize in ways that don’t directly make demands of our customers. It would be possible to produce very sophisticated market data from the information our service will collect, something that our partners are likely to find very valuable.