Thanks for your insights. Still trying to decide whether something like this should go on my wish list ;) (see above for my potential uses).
Don't expect to much of it for the first generation / first product. Once you get a few competitors and iterations on the technology, you can have a serious look at it. I'd also wait to see a view "in-depth technical reviews" with labor-based comparison studies. To satisfy your curiosity, if your budget allows it, it will definitely be worth it (assuming it at least work a little). I expect the results at first to feel a little like translating something in some language with google translate and translating it back with bing.
How accurate, exactly, do you think such a device could be? Obviously it's not going to be pulling out the sort of precision of a professional spectrometer. But you mention, for example, being able to identify the signatures of herbicides and pesticides. Do you mean, for example, "This contains imidacloprid", or more like, "This contains a nicotinoid of some variety"?
It's hard to say without having more detailed information on the spectral range, spectral resolution, illumination source, detector type and sensitivity, and, most importantly, the type of spectrometer. From the description, it seems that it is a diffracting spectrometer, but that again comes in different flavours. Long story short, I can only guestimate the performance of it based on my experience designing such devices with the information they provide. Now, some molecular basis have specific signatures even a low resolution device could identify. Because these molecular basis are shared by various different molecules, it may be difficult to do more as to identify a group such as nicotinoids. And for the same reason you will have a lot of false positives.
This is where the apriori information play an important role. If you are looking at an apple, the spectral database-based/cloud-based analysis program knows what to expect and can raise a warning flag if it sees something which might be a nicotinoid, because that's not expected in the spectra of a "clean" apple. The same analysis program will ignore the warning if you say you are analysing a cigarette. This, btw. may also lead to false negative... but with low spectral resolution, I believe that the cloud-based analysis will play a crucial role. And this is actually the most interesting part of this innovation.
How useful do you think it could be on identifying mineral species - say, distinguishing between different zeolites?
I've never saw an IR spectrometer used for mineralogy or metallurgy. I doubt it can provide you with any significant information in those fields. Other factors you cannot see in vibrational spectroscopy play a too large role.
Or, back to food, if given, say, a mango, to get readings of, say, water, sugar (in general, or specific sugars), fat (in general, or specific categories of fats, or specific fats), protein (in general, or specific categories of proteins, or specific common protiens... obviously it's not going to be able to pull out 5 ppb of Some-Complex-Unique-Protein), common vitamins (generally found in dozens of ppm quantity - some more, some less), minerals (likewise), etc?
This is where I believe the advertisement is way to promising and optimistic. I'd occupy your day to do that on a very expensive and high-performance laboratory IR spectrometer. Under laboratory conditions.
Take this example of sugar spectra from the Agilent website. They don't specify it, but that's most likely 4 cm^-1 resolution. You see that it is possible to identify specific sugar compounds, but also that the spectra are quite alike. Now reduce the spectral resolution to something like 128 cm^-1 and you'll have a hard time identifying the exact type of sugar.