Your first order of business before reading my take should be to read DAGAP. It is still very relevant and gives a good high level overview of what's expected at top schools. I will only propose changes to some parts and provide some timelines / strategies I used. For some context, I was a part of the Admissions Committee at CMU for MSML. While I won't talk about any of the specifics, I do think I observed some high level changes from DAGAP and it is OK to talk about them. I also have some insights from my friend's and my outcomes.
Note that admissions into top schools are very stochastic, and I would recommend applying to multiple programs even if you are confident about your application. This post is geared towards international students applying to US schools. It's not relevant for any US nationals.
When I applied to universities for Fall'25, most universities did not consider the GRE or had made it optional. It is also optional for MSML now. I personally did not send it to schools where it was optional. It costs a lot of money to give the GRE and also send it to multiple schools, so before taking it, do check if you even need it.
If you still have to submit it somewhere, a score of 320 and above is adequate. A higher score is better, but it will never be the reason for your acceptance. If you have a very low score though, it can be a reason for rejection. The quant score should be greater than or equal to 168. The verbal score can be compensated with an excellent TOEFL / IELTS score. I myself had a GRE score of 323 (Q: 168, V: 155, A: 4.5), I did submit it to CMU's MSML program. I know a friend who scored 321, and got admits from top schools like UMich and Georgia Tech. Thus my opinion, it will never materially change your result for good.
For TOEFL, I just have a minor addition to DAGAP, some shools set up a sectional minimum score as well, and in extreme cases, would not look at applications with less than the sectional minimum. For example, Princeton requires a speaking score of at least 28, I have not seen many other schools who set up such stringent criteria. A score of 110 is very safe, but over 100 shouldn't be a problem if it's distributed well across the sections. I would say that having a very low score in any section is always flagged, and can be a concern. I scored a 114 and this is usually mandatory to submit. A bit of BITS specific advice: BITS has some partenership with ETS and it gives students 8 free sends. There are 4 free sends anyways, so you can apply to 12 schools without worrying about the cost of sending the scores.
Preparation -- it's a whole different article, as a rough estimate, I prepared for the GRE for about 3 weeks and for TOEFL, only a day. Feel free to email me for details. Finally, you should plan to give both of these at least about a month before the deadlines, to have results in time.
I strongly agree with DAGAP. It is important to take difficult courses and do well on them. Also as an additional note, since it is very hard to calibrate performance across schools, it really helps if some recommender can comment on the quality of students and the rigor of the courses at your institute; unless it's very well known. It is fair to request this from your recommenders :)
I would advise anyone to put their primary focus on their academics, and raise their GPA from whatever it is. Investing time in academics is a low risk high reward strategy. Trying to compensate GPA with any other aspect is a high risk high reward strategy, it only works sometimes.
Again, I agree with DAGAP entirely; though due to that, I didn't pay a lot of attention to my SoP, and wrote it much later (only a few days before the deadline). I would advise anyone to start much earlier (about 2 weeks)! I am also sharing a few resources I used, [MIT CommLab][example]. I would also suggest too get it reviewed by as many people as possible, especially those who have been through the process (MS / PhD applications). You should write it once before any review / examples so you're not biased towards a style or flavour.
Finally, I shortlisted universities based only on some rankings and prestige, thus I would also go through the entire faculty list of the department manually and mention professors I liked and would want to work with, in most cases there would be muiltiple natural alignments.
I strongly believe that after GPA, this is the field that is given the most importance, as this is entirely out of the applicant's control. I think DAGAP places most of its opinions correctly, I'm not sure about the negative points part, overly glorifying letters are definitely bad.
As a strategy, reach out to professors much earlier, about 7-8 weeks before the deadlines. Most professors would ask some summary / bullet points of their experience with you, otherwise as well, it would not hurt to send such a summary along with your updated CV and transcript. Also don't shy away from asking to highlight specific details, like TA-ship experience, etc.
This was probably not a major factor in decisions at the time of DAGAP I think (it is a subsection not a section). However, the landscape has definitely shifted over the years. Much more for AI related fields, but also for other subfields in CS, like networks or systems1. I think even then, the guidelines there are still correct.
I think some form of research experience is now an unwritten requirement for most top schools. Note that I am saying research experience and not publications (I strongly believe undergraduate research is really random, but that's another story). I think so because the applicant pool is so strong, and the seats so few, that the committee will always have enough applicants with research experience.
Before I get cancelled by the research community, you should not do research in order to get into a top school. What I mean is, if your primary goal with research is to get an admit, you're not doing it right. Your primary goal with research should be to do research. Research strategy is an article in itself, though I would suggest to start early, as early as possible.
The advice in DAGAP is again quite apt, as a side note, I have seen people applying usually with 2 years of experience. I would strongly advise anyone taking this path to get into some top company or work in a top research lab / startup. For India, Microsoft Research and Google DeepMind have their own (competitive) pre-doctoral programs [MSR] [GDM] GDM posts the opening when its open. Top IIT / IISc professors are also more than willing to host students for a year or so.