Akuna Capital · Phone screen → Onsite
In processAkuna Capital JQR CodePair Rounds: 8 Experiences (Part 6 of 6)
Akuna Capital quant researcher interviews from 2017–2018: eight reports with structured matrices, conditional normals and statistics-class coding.
At a glance
- Process (2017–2018): one or two phones, sometimes a CodePair class-design task, then onsite coding and derivations (H68–H75).
- Asked in these reports: probability in 6 of 8, linear algebra and coding in 5 each; repeats include a structured-matrix question (H73, H75) and statistics classes coded live (H69, H70).
- Difficulty: one JQR phone was very short, a single conditional-normal question (H71).
Coverage: 8 Akuna Capital quantitative research reports from October 2017 to March 2018 (QR intern, QR, Junior QR, ML track). All are more than three years old, so they sit in History below.
In this series
- Part 1: 2022–2025 screens, coding and the newest reports.
- Part 2: 2021–2022: coding, simulation and probability.
- Part 3: 2020: root finding, regression and shuffles.
- Part 4: 2019–2020: ML and final rounds.
- Part 5: 2018–2019: probability and SVMs.
- Part 6 (current): 2017–2018: JQR, ML and projects.
Newer reports: Akuna quant technical interviews, 2025–2026.
What came up
Linear algebra and live class coding stood out in this period, next to the usual probability questions. Every question stays in its entry.
Common questions
What coding did Akuna ask quant researchers in 2017?
Live statistics or real-time classes on CodePair (H69, H70, H73) and onsite coding with distribution fitting (H74).
History (before 2024)
H68: Two phones and finals with derivations and data structures
Report date: 2018-03-06 · Final interview: 2018-03-02, inferred from “last Friday”; phone dates not supplied · Role: Quantitative Research Intern, ML
The mathematics phone was scheduled for 45 minutes and finished in 30; the next phone was scheduled for 30 and took 35. The final web interview had three interviewers and ran beyond its 90-minute slot after connection trouble. Rejection arrived an hour later.
Mathematics phone
- Explain uncorrelated versus independent variables and give a counterexample.
- When can a limit and an integral be interchanged?
- Find the covariance of Brownian motion at times s and t.
- Discuss fixed-point iteration. The candidate described it as a fixed-point theorem without identifying a particular theorem.
- Compare almost-sure convergence with convergence in probability.
- SVD and other matrix decompositions were discussed.
- Projection matrices also came up.
- The interviewer mentioned the kernel trick.
- Monte Carlo integration was another topic.
- Lagrange multipliers were discussed.
Python and data-structure phone
- The Python phone covered class inheritance.
- It also covered the generator keyword.
- List comprehensions came up, without a retained expression.
- The candidate clarified that garbage collection meant Python's basic GC mechanism.
- Explain how to maintain a data-stream median using two heaps. This was an oral algorithm question; coding was not required.
- There were further Python and data-structure concepts that the candidate could not individually recall.
Final: calculus
- Derive the derivative of arctan(x), rather than just quoting it.
Final: ML
- For a binary-classification problem, does a model with 95% accuracy necessarily exist if unlimited training data are available? The interviewer brought up Neyman–Pearson; the distributions were not given.
Final: project and problem solving
- Explain a project in terms an undergraduate could understand.
- Count integers from 1 to 10,000 that contain the digit 5.
- Two players each toss their own coin at the same time. One is waiting for HH and the other for HT. Which has the better chance of winning? Tie handling and the reset convention were not recorded.
- Deduplicate with a set or dictionary, explain O(n) time and storage proportional to the unique values, then optimize for a very long list drawn from a small known universe. The candidate considered early stopping; the account does not settle whether that was the intended solution or an output-order requirement.
H69: A matrix parameter, a conditional density and a statistics class
Report date: 2018-01-29 · Role: Quantitative Researcher
There were two 30-minute phones: mathematics first, then Python CodePair and brainteasers.
First phone
- For independent standard-normal X and Y, derive the density of X conditional on X+Y>0.
- Find eigenvalues and eigenvectors of a matrix with ρ on the diagonal and 1 in every other entry.
Second phone
- Remove duplicates.
- Implement a list-based class that returns the mean, maximum and mode.
- Write a list comprehension; the expression was not retained.
- How many integers from 1 to 10,000 contain the digit 5?
- Distribute 50 white and 50 black balls between two randomly selected baskets to maximize the chance of drawing black. The follow-up asked for intuition beyond the optimization calculation.
H70: Mathematical concepts followed by a CodePair statistics class
Report dates: 2017-11-14, 2017-11-30 · Role: Quantitative Research Intern, ML
The first conversation covered mathematical concepts for the Summer 2018 research internship.
The later CodePair contained three tasks.
Mathematics phone
- Newton's method was part of the mathematical-concept discussion.
- Lagrange multipliers were discussed.
- Martingales came up.
- The interviewer also asked about Brownian motion.
- Poisson processes were another topic.
- Lévy processes were discussed as well.
Later CodePair
- Remove duplicates without changing the list's order.
- Use a list comprehension to produce powers of 2 between 2 and 16.
- Implement insert(x), get_max, get_mean and get_mode. Optimize query time; insertion time was not important. The candidate got stuck on the mode query and eventually proposed two dictionaries. No exact constant-time requirement was stated.
H71: A conditional-normal question in a very short JQR phone
Report date: 2017-11-21 · Role: Junior Quantitative Researcher, ML
After a project introduction, the interviewer asked one mathematical question. The entire call ended in less than three minutes when the candidate could not answer it.
Technical phone
- For standard-normal X and Y, find the conditional distribution of X given X+Y>0. The account did not state independence. This one question brought the interview to an end in under three minutes.
H72: Regression assumptions under correlated or nonnormal data
Report date: 2017-11-18 · Role: Quantitative Researcher, ML
The candidate confirmed a phone-only conversation without CodePair. Rejection was communicated three days later.
Technical phone
- Explain independence and zero correlation, including an uncorrelated-but-dependent example. If both variables are Gaussian and uncorrelated, are they independent? The question as written did not distinguish marginal Gaussianity from a jointly Gaussian pair.
- State linear-regression assumptions. What changes when errors are nonnormal, and does the regression itself stop working?
- How would you handle highly correlated features? What does PCA do, and what other dimensionality-reduction methods could you use? The candidate confirmed that this was a phone discussion without CodePair.
H73: JQR: a structured matrix and a real-time class
Report date: 2017-11-07 · Role: Junior Quantitative Researcher
A mathematical phone was followed by CodePair and brainteasers, lasting more than 30 minutes. The candidate was waiting to hear about the next step.
First phone
- Why Akuna?
- Find eigenvalues and eigenvectors of an n×n matrix with 1 on its diagonal and n in every off-diagonal entry. A later clarification confirmed this layout.
- For independent standard-normal X and Y, find the conditional distribution of X given X+Y>0.
Second phone
- Delete repeated values from a list.
- Use a list comprehension to generate powers of 2 between 2 and 16.
- Build a real-time class supporting insertion, maximum, mean and what the candidate called the major element.
- How many integers from 1 to 10,000 do not contain the digit 5?
- Allocate 50 black and 50 white balls between two baskets to maximize the chance of drawing a black ball.
H74: Two phones, then onsite coding and distribution fitting
Report dates: 2017-10-27, 2017-11-06 · Role: Quantitative Researcher
Two phones covered basic mathematics and coding. An onsite followed.
The Chicago onsite included HR, mathematics/statistics, coding and a conversation with the head of quant. An ML round was described as optional.
Mathematics phone
- The mathematics phone ranged over probability and measure theory, stochastic processes, linear algebra, polynomials, ML and statistics.
Coding phone
- The coding phone involved a straightforward algorithm task or tasks. The candidate did not retain the problems or their count and said they were not asked to write classes.
- Generators and list comprehensions were also confirmed, without the particular expressions.
Onsite interviews
- The onsite included mathematics and statistics, without retained individual questions.
- There was a coding round, but its task was not identified.
- The head of quant asked an elementary geometry question; the figure and measurements were not retained.
- Given data, fit its distribution and implement the fitting algorithm yourself. Pandas was allowed, but a ready-made package for the actual fitting was not. The family, criterion and required output were not recorded.
H75: One structured-matrix question for a research internship
Report date: 2017-10-13 · Role: Quantitative Research Intern, ML
The candidate twice confirmed that the structured-matrix problem was the only question.
Technical phone
- Find the eigenvalues of an n×n matrix with n on the diagonal and 1 everywhere else. The candidate twice confirmed that this was the only question.
Keep practicing
- From these reports: maintain a data-stream median using two heaps (H68), remove duplicates preserving order (H68), stream max mean mode (H69), powers of two in range (H70), handle highly correlated features (H72).
- Akuna Capital question banks: algorithms (sign-in needed).
- Everything Akuna Capital in one place: the Akuna Capital interview page.
- More firsthand reports: the Experience feed.
In this series
Sources: Anywhere useful interview info can be found.
Comments
0No comments yet — be the first to share what you know.