I did my BTech in Civil Engineering in 2025 and wrote GATE CS 2026. I decided to opt for MS programs in top IITs but unfortunately did not get call backs.
Luckily I wrote PGEE exam and did reasonably well I guess (They dont reveal the marks for MS students) so I was invited for MS interviews.
At IIITH the approach is different from that of other institutes. It is comprised of multiple research centres instead of a single CS department and these centres take the interview. I got invited to interview for my top 2 centre preferences, CSTAR (Centre for security, theory and algorithmis research) and MLL (Machine Learning lab)
1st Interview was MLL.
Panel:
P1:
P2:
P3:
The main lead was P1
Me: Entered. They did not tell to sit. Told me to erase blackboard (:P)
P2: (He was reading my details) When did you graduate?
Me: 2025
P2: (Smiling and looking at others) Is there a problem in the form? It shows his graduation date as 2003
Me: Sir, must be a glitch (Smiling sheepishly)
P2: I believe you, you dont look like you are 46 years old. (Corrects the date on the application)
P2: Where did you do your bachelors?
Me: I did my BTech in Civil Engineering from XYZ college. (P2's and other profs eyebrows raised when they heard civil engineering)
P2: So how much did you score in PGEE?
Me: I am not aware sir, the marks are not revealed to me.
P2: (Looks down, the marks were listed) You have got a very good score considering your background. Did you take any courses in CS in college?
Me: Yes sir, also I prepared for GATE for 6 months. Told all the courses i took.
P2: Interesting, what was your gate score and rank?
Me: 644 and 209X
P2: Thats a very nice rank. (I smiled and said thank you sir)
Meanwhile P1 was reading my SOP and resume to find the best area to roast me on ;)
P1: I see you have written you like graph theory in SOP. Can you tell me what is a matching?
Me: (stuttering)... Sir informally..
P1: (Smiles) this is a formal interview
P1: Can you write it on the board?
Me: I write the definition on the board, and they seem satisfied.
P1: Define a maximum matching
Me: Defined correctly.
P1: If we have a connected graph, what is the lower bound and upper bound on Maximum matching?
Me: (I take some time to think) sir in the case of lower bound (I said lower bound by mistake) each edge in the matching will match one pair of vertices, so maximum we can have n/2 such edges.
P1: isnt that upper bound
Me: yes sir, sorry i meant upper bound.
P1: ok now tell me the lower bound.
Me: (I start drawing some examples as I was not sure of the answer. I drew a cycle first) Told the maximum matching in a cycle which is a special case of connected graph
P1: that is correct but this is special case... tell me the general case
Me: (I was still not sure, this time I drew a complete graph, then I figured out that complete graph actually is the upper bound example) Sir in the complete graph we get the upper bound
P1: yes I know that (Little annoyed) Tell me how to get the lower bound
Me: (Now I started drawing a path graph... then I saw the connection...) Sir to this path graph if I keep adding edges to a vertex the matching size does not increase
P1: you are close.. complete the argument
Me: (Finally figured it out). Yes sir if I just add edges to a single vertex, the maximum matching will always remain 1 so lower bound is 1.
P1: Correct, which graph is this?
Me: Stuttering (Lol I cant believe I forgot the name of the graph)
P1: (Smiling) Isnt this the star graph?
Me: yes sir yes sir (Smiles awkwardly)
P1: ok now define an independent set
Me: Defined correctly.
P1: Draw an example and label the independent set.
Me: Drew correctly.
P1: Do you know what is a clique?
Me: Defined correctly.
P1: Tell me the connection between an independent set and a clique
Me: I said I am not sure.
P1: Do you know complement of graph
Me: Yes sir
P1: That is your hint.
Me: (Flustered again) I start drawing some more examples to buy time...
Me: Complete graph is also an example of a clique... if I take complement of a complete graph I get the empty graph which naturally is an independent set.
P1: Close, finish the argument
Me: (Realising) An independent set in a graph G is a clique in the complement G'.
Then the panel went to some math question which was very weird and cant really be explained here...
P3 finally speaks:
P3: Do you know algorithms?
Me: yes sir, I studied in GATE
P3: No I mean, have you implemented any classical algorithms like sorting, searching in code? Like in college?
Me: No sir, I have mostly done coding for software projects (Civil background)
P3: (Confused)... ok you can go now
P2: (Addressing P3) I think we can ask him a programming question
P3 (Thinks for a while): Ok write a code to reverse string
Me: Writes code correctly, P3 looks satisfied.
P3: Ok now you can leave. Take your documents.
This was a very weird experience because they grilled me on a concept in graph theory which even CS students dont study thoroughly (let alone civil students), and this was an ML lab so I expected ML questions or linear algebra (Had prepared more for that). I am not sure of selection.
Also they did not ask me anything about my resume, projects. They also seemed very strict.
Next interview was of CSTAR
Panel:
P1:
P2:
This time the interview felt more comfortable. They asked me to sit.
P1: Introduce yourself
Me: introduced.. again got same reaction when I said civil engineering.
P1: What subjects you are comfortable with. (Atleast he asked)
Me: Sir, I like linear algebra, algorithms...
P1: Ok you can go to the whiteboard.. he tells me to draw a 2*2 matrix.
Me: Drew
P1: Now lets say this matrix is M. How to find M^10?
P2: jokes.. why are you telling him to find M^10, let him find M^100 or M^1000
Me: (Sweating) Sir.. I think we can find eigenvalues of this matrix.. (I calculate eigenvalues)
P1: Ok...
Me: Then we know that if the eigenvalues of M is a,b then eigenvalues of M^10 are a^10, b^10
P1: That is correct. How to use that to find M^10?
Me: (I was stuck here as I did not know diagonalisation) I drew a general 2*2 matrix with 4 unknowns... then I wrote out the characteristic equation and plugged each eigenvalue which was calculated from M
P1: Interesting approach... but wont you get only 2 equations? How will you solve for 4 unknowns? (I think he was testing me because he had purposely given a symmetric matrix which I did not notice. In that there are only 2 unknowns, not 4)
Me: (I was stumped here)...
P2: Do you know diagonalisation?
Me: no sir.
P2: Ok... then leave this approach.
P1: Still, its a nice approach, try doing it later
P2: Tell me if you just had to quickly solve M^10 in an exam what will you do?
Me: Sir I can repeatedly square it.. wrote M^10 = (M^2)^2^2 * M^2
P2: So how many operations?
Me: 4
P2: Ok tell me another subject
Me: Sir, graph theory and algorithms
P2: Define dijsktra algorithm
Me: Told
P2: Does it run with negative weight edges.
Me: Sir, while the algorithm is not intended for negative weight edges, there are some special cases where negative weight edges will still yield a correct answer.
P2: interesting. can you draw an example
Me: (Struggling to draw)
P2: (Gives a hint, kind of). Draw any graph which has -ve edges.. show me dijkstra on it
Me: Draws and shows. then realises that graph does not terminate if the -ve edge is on the shortest path
me: Sir algorithm will not terminate as the negative edge is on the path.
P2: yes, (Then he explains a full story about how shortest walks itself are ill defined and that is why we cannot solve for negative edges using Dijkstra)
P2: Now what about MST. Define MST.
Me: Defines.
P2: Can we run any MST algorithm on -ve edges?
Me: No sir. (This is wrong answer but for some reason he continued and did not correct me)
P2: Ok now let me define a maximum spanning tree as the spanning tree with maximum cost of edges. I want to find the maximum spanning tree of the graph but I cannot write my own algorithm, I can only use the traditional MST algorithm like Prim/kruskal. The only thing I can do is modify the weights of edges.
P2: (Continues). Naturally one would think to invert the edges and make them negative but you just said MST algorithm cannot run on -ve edges. So what to do?
Me: (Stumped) Draws some example to buy time.
P2: Think of K's complement
Me: it is some w - number.
P2: Ok so how do we invert the edges but keep them positive so the MST algorithm can work
P2: (He does not wait for my answer) we can just take max of all edges and subtract the weight from it right?
Me: I nod ;)
P2: ok good. Tell me one more subject
Me: Sir, I am good with algorithms
P2: Ok do you know recurrences?
Me: i can try my best to solve sir
P2: Ok write a(n) = a(n-1) + a(n-2) does this look familiar
Me: Yes, it is fibonacci recurrence
P2: Ok solve and find closed form.
Me: Solved.
P2: Ok come and sit
Then the panel grills me on my personal motivations for more than 15 mins which I wont talk about here.
Each interview felt like almost an hour. I dont know if it is because I I am dumb but the questions were quite difficult to me and required me to think a lot, most other people did not have the same experience as their panels focused on the projects and resume more. (I have a data science internship so i was hoping for questions on that). I also did not expect graph theory in an ML lab interview (,,>﹏<,,)
Results are not out yet. I am not very hopeful of selection tbvh.
Edit: I ended up getting selected for CSTAR :))
