That's exactlty why suxh a talk would have been a good idea...

If that is what the subject of the talk was then I completely agree it would have been appropriate and rescind whatever negative thought I expressed in the parent. (And by "that" I mean "A practical guide to detecting caste-based prejudice and what to do about it.")

Care to elaborate?

It's this naive idea that somehow same tactics which have clearly failed in regards to white/black race relations, would work on a totally different culture. And yes, I know you may think that this woman giving this talk would be someone who knows a thing or two about the Indian culture, but you would be wrong. All she knows is how to sell an offering which western executives would pay for.

Lemme put it this way, if the solution required to fix the problem could be one of the following, which one do you think you would go for:

1. Performing caste cleansing rituals throughout all teams

2. Hiring a caste system expert to give a talk in the company

3. Identifying the troublesome castes by their surnames and making special reports on managers with that caste names regarding any issues of caste discrimination.

Which one of these options would you go for?

Even if the effectiveness of the option 2 is the least, it still would be the most preferred option by all companies because it's the best bang for their buck. In fact, even if these talks were actually toxic for the morale of the company among the Indian diaspora in your company, this gives the most 'color' of action by the company (and when companies like Google resist as in this case, they look bad).

End result: When you need a quick and expedient action to be performed in case of dealing with an issue you have no real understanding of, you are bound to hire scammers whose main job is to give you a totally fake service with no outcome being achieved.

how does it work with gender/race? If your white co-worker likes a candidate, it is because of his gender, race or actual competence?

I have been a part of something like 150 interview loops. In that time I think I've interviewed exactly 1 white woman. In her case, everyone but a single person was in favor of hiring, and that one holdout was adamantly opposed (and eventually won). The person against hiring never said "I hate her because she's a white woman" but instead had reasonable sounding issues.

Given that "white woman" is such a rare category to interview, and, I assume, Dalit is pretty rare as well, AND I don't even regularly interview with the same set of people - how could I possibly tell when another interviewer is motivated by prejudice? Unless the interviewer says they are motivated by prejudice I'm either going to have to give them the benefit of the doubt or make a heckuva leap to accuse them of something.

I think in part because there is direct legislature that makes certain hiring practices illegal and cause for retaliation (obviously this is fraught with problems in proving intent like everything, but if I'm from a particular ethnicity and I have a track record of dismissing candidates that are able to be isolated down to a protected trait rather than ability to do the job, then I can see that as being relatively obvious.) but Like i said it's fraught with external factors, but as far as I know a caste isn't considered a protected class (probably because it doesn't have strict parallels in the West).

I'm familiar with the broader social context of gender and race in the US. So when I see, for example, a coworker who consistently argues that female candidates are too aggressive, I can gently prod the discussion towards focusing on his technical feedback and deferring to the other interviewers for soft skills. With caste discrimination, I have absolutely no idea what the equivalent would be, and it'd surely be inappropriate to corner one of my Indian coworkers in the break room and ask her to teach me all the common Dalit stereotypes.

all of your questions are precisely why this needs to be discussed—interviewers clearly need a better grasp on the subject.