All the pre-processes were held offline on our campus @IIT Delhi.
First stage was résumé screening, where one of the criterias was a minimum CGPA of 8.0/10

Second stage was an OA (Online Assesment), which was monitored in real-time both by invigilators and webcams. It was a coding test of 60 minutes, on the platform HackerRank. So it is advised to get comfortable with working on an online IDE, during regular practice.
There were two questions which were asked:
One was a simple application of the Knapsack Problem, with memory optimization. The other was almost a LeetCode medium to hard level greedy problem, on which I had used Binary Search.
Some similar problems, along with their variants:
Knapsack 1 (Classic Knapsack)
Knapsack 2 (Unbounded Knapsack)
Maximize the Minimum Powered city (Not the exact question but just to give an idea of the difficulty)

Final Interview
After the previous two, I was luckily (nothing is luck) shortlisted for the interview process. It had two technical rounds, which was followed by an HR round. The details are as follows:
1st Round : After a short intro, I was asked to explain a project which I mentioned in my portfolio. It was a 2-D Range Searching problem which I had worked upon. Here is my GitHub repo on the same - Search Nearby Restaurants. After that, I was given this standard problem on Boundary Traversal of a Binary Tree (you can find the free version on GFG). This is a medium-level problem on GFG and LeetCode. Though I intially did miss an edge-case here, I found it later and corrected it.

2nd Round : Here, I was presented with a LeetCode hard-level problem. I had to design the Uber Pool application, where I had to provide the end-to-end thought process along with the Optimization requirements. I'll not discuss the entire detailed solution over here, but rather write about the overall approach, which I came up with.
First, we shortlist the possible cabs, which can be allotted to a person who has currently booked. This will be based on various parameters, ranging from cab and driver info, expected time required to reach the destination etc. Simultaneously we also create a graph with appropriate weights.
Note : The weights is a bit tricky, since you have to take care of both the fuel consumption of the car, as well as the passenger's time. So some weighted mean? But the units are different! Do think on this!
Now, let's assume that the cab is currently full, ie, there are 4 dropoffs to be made. First, simply precompute the best possible path between every combination of two particular points, using an appropriate algo. Then, you can use a brute force method (this is not really costly since there are only 24 options) to determine which is the best possible sequence of dropoffs. Also note that, between any two consecutive dropoff points, you will be following the optimal path, which you previously computed.

Final HR Round : If you are here, most probably you are through!
This was mainly a behavorial round, where I was first asked to give a short intro. This was followed by some questions about how I was different in my Course Projects, and whether I had worked on projects outside the general coursework. Some basic intelligence-testing questions were also asked. In short, don't try to ever brag about yourself too much, since the interviewer will ofc be much smarter than you. If you are asked about your positives and negatives, speak realistically while being grounded, else it might be considered as a negative.

So, this was all from my side about my experience!
And about the practice suggestions, all that I've got to say is that practice the three topics - Greedy, DP, Graphs very thoroughly. The best strategy as per me is that, put these topic filters on LeetCode and solve (every medium and hard problem) as much as you can.
To know where you stand, you can give contests on Codeforces and Atcoder on a regular basis.

And finally,
Good luck with whatever you are up with!