This time to share my experience in Virtu Financial ‘s interview process lasted about two weeks, from the relaxed OA to the math-filled Final, which was not fast-paced, but the experience was really "highs and lows", especially in the final round, which was a bit hard to describe.
OA Session: Regular Topics, Easy Pass
Virtu's OA belongs to the set that has been widely circulated on the internet, which is not too difficult, and the question types are relatively fixed. I've read the question bank in advance, so there's basically no surprise when I do it. The test platform was smooth, and I submitted it in less than half an hour after finishing it, and the result came out very quickly, and I passed the test successfully. On the whole, OA is the easiest part of Virtu's interview process.
Technical Interview: Interesting Black Box Function Questions
This round of Tech interview experience is actually quite good, the other interviewer is very professional but friendly, mainly examining the thinking logic and algorithmic thinking ability.
The gist of the title is as follows:
Given a function f(x) with the domain of definition [0,1]. It is known that f(x) has a unique minimum in the interval, but the function is not derivable and has no expression (treat it as a black-box function). Ask if you have an algorithm that can estimate the point of the minimum, which is required to satisfy a certain degree of accuracy.
The idea of the solution is to use an idea similar to "trisection", divide the interval [0,1] into three segments, take two equal points a and b, and compare the values of f(a) and f(b):
- If f(a) > f(b), the minimum must be at [a,1];
- in [0,b] if f(a) < f(b);
- between [a,b] if f(a) ≈ f(b).
Repeat this process until the interval is small enough. This is a typical optimization problem, where the focus is on how you use the "comparison results" to shrink the search interval.
After this round of interviews, I felt in good shape, the pace of the 40-minute discussion was smooth, and the communication with the interviewer was also quite smooth, and the other party also said that "the idea is quite clear" at the end.
Final Interview: from research to mathematical modeling, the experience went downhill!
Final Two rounds in all, and all I can say is that the experience was very mixed.
first round (of match, or election)
As usual, the opening session was a background chat, mainly about the research and projects I had done, and the other person listened quite attentively and asked detailed questions, so the overall atmosphere was good.
Then it started getting into the math:
How to generate a 2D uniform distribution using a 1D [0,1] uniform distribution?
If the 2D space is of arbitrary shape (e.g., triangles, convex polygons, etc.), how to generate uniformly distributed points?
This problem is actually a pretty interesting probabilistic modeling problem. Ideas can start from mapping:
- The two-dimensional square [0,1] × [0,1] is first constructed using the uniform distribution of [0,1];
- A mapping is then used to map the boundary of the square to the target convex region;
- The interior points can be approximated by the weighted average of the boundary points.
I was sketching as I spoke, the other person nodded in approval, and overall it felt pretty smooth.
second round (of match, or election)
Continues to talk from the background, but quickly switches to more theoretically oriented topics:
Initially A=1, B=1.
One is randomly selected in each round plus 1, and the probability of being selected is proportional to its current value (the probability of A is A/(A+B) and the probability of B is B/(A+B)).
Q: What happens after multiple rounds?
I made a list of the first three rounds and found that the sum of A+B was always increasing, but the probability of each possible distribution of A and B was equal under the fixed sum condition, that is, a relatively "smooth" process.
The questions are not particularly difficult, but the interviewer's style of questioning has become a bit "dogmatic", almost not much guidance, so that you unilaterally deduce, the experience is really down a lot. The moment I finished the interview, I felt that this round might be hanging in the balance.
Overall impression
Virtu's process is quite standard: OA → Tech → Final, and the results will come out in about two weeks. The questions in the Final stage are not particularly tricky, but they are more oriented to theoretical deduction and require a high communication pace.
If you are a student preparing to pitch for a Quant or Algo related position, it is recommended to focus on preparing for mathematical intuition and probability derivation type questions such as:
- The class of function optimum estimation (optimization intuition)
- Stochastic modeling (stochastic process modeling)
- Distribution mapping & transformation (sampling & transformation)
Unlock the big factory Offer, PROGRAMHELP full escort!
Imagine, from OA written test to interview VO, to signing negotiation, all the way to FAANG Offer without worry! Our PROGRAMHELP team, consisting of 7 top IT elites: 4 graduated from Oxford University, Princeton University and Peking University Masters and PhD, and the remaining 3 are now working at Amazon, Google and Ali tier 1 big factories. The founder himself graduated from Peking University and has 10 years of experience as a first-line engineer, and is personally responsible for - resume packaging, interviewing and code writing,OA/VO Assist, whole process is one-on-one communication, no intermediary spread, and direct to effectiveness!
