Akuna Capital · Phone screen → Onsite
In processAkuna Capital QR Coding Interviews: 17 Experiences (Part 2 of 6)
Akuna Capital quant researcher and intern technical interviews from 2021–2022: 17 reports heavy on coding, probability and numerical methods.
At a glance
- Process (2021–2022): one or two technical phones, often on CodePair, then final rounds mixing math, coding and research talk (H15–H31).
- Asked in these reports: coding and data structures in 12 of 17, probability in 7, ML in 4; repeats include the maximum-product subarray (H27, H29, H30) and numerical integration (H21, H22).
- Difficulty: one JQR phone in Europe was all probability with no coding (H17); others expected working code within a phone.
Coverage: 17 Akuna Capital quantitative research reports from January 2021 to February 2022 (QR intern, QR, Junior QR; Chicago, Boston, Shanghai and remote Europe where stated). 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 (current): 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: 2017–2018: JQR, ML and projects.
Newer reports: Akuna quant technical interviews, 2025–2026.
What came up
Coding carried more weight here than in the older parts: subarray products, time-window queries and integration calculators, alongside classic probability puzzles. Because every report is older than three years, there are no current tables; each question stays in its entry.
Common questions
Does the Akuna quant research interview include coding?
In this period, usually yes: 12 of 17 reports had coding or data structures, such as a maximum-product subarray (H27, H29, H30) or an integration calculator (H21).
What probability questions did Akuna ask quant researchers in 2021?
Uniform sums, absorbing walks and random hats (H23), parity of heads and increasing random sequences (H24), among 7 of 17 reports with probability.
History (before 2024)
H15: Model metrics, betting and a rotated-array problem
Report date: 2022-02-28 · Role: Quantitative Research Intern
The US internship phone lasted about 30 minutes and combined statistics, probability and CodePair, with the camera off. Only about ten minutes remained for the coding explanation. The interview was Monday afternoon; the rejection arrived Thursday morning.
Technical phone
- Explain R².
- Explain the bias–variance tradeoff.
- Start with a balance of 100 and continue until reaching 500 or going broke, with win/loss probabilities 0.4 and 0.6. Compare all-in, half-balance and fixed-10 policies. The candidate interpreted the goal as maximizing the probability of reaching 500.
- Search in Rotated Sorted Array came up on CodePair. The candidate said roughly ten minutes remained, enough to explain the idea.
Also discussed in the phone
- The candidate also recalled a project discussion covering technical details, a challenge and how they solved it.
H16: A short resume discussion and one coding task
Report date: 2022-02-24 · Role: Quantitative Research Intern
Resume discussion and one coding task took less than 30 minutes.
Technical phone
- The interview combined resume discussion with one coding question. The candidate did not name the task; the whole conversation took less than 30 minutes.
H17: A remote JQR probability interview with no coding
Interview window: 2021-12-06 to 2022-02-01 · Role: Junior Quantitative Researcher · Location: Remote, Europe
The candidate applied on December 6, 2021 and was rejected on February 1, 2022. A quant-team interviewer used CodePair. Coding had been announced, but no coding question was actually given.
Technical interview
- Solve a probability problem involving expectations. The candidate kept trying conditional expectation and had trouble following the hints. They deliberately did not share the setup.
- The interviewer also discussed the candidate's CV and project.
H18: Research, time-window queries and numerical stability
Report date: 2021-12-13 · Role: Quantitative Researcher; seniority not stated
About 15 minutes with HR preceded three 45-minute final interviews. Each of the first two technical rounds spent about 20 minutes on research before moving to a task.
Final interview 1
- The first final-round conversation began with about 20 minutes on the candidate's research.
- Implement an observation class: on_observation stores a timestamp and value, and get_max(timestamp, window) returns the largest observation strictly inside (timestamp − window, timestamp). Both endpoints were excluded.
Final interview 2
- Another interviewer spent about 20 minutes on research before coding.
- Use gradient descent to find the positive solution of x^N = a, where a > 0 and N is a positive integer.
Final interview 3
- When is XᵀX noninvertible in linear regression? Why does Ridge help?
- Compare Newton's method with gradient descent, including their advantages and drawbacks.
- There was an order-book analysis question, without enough detail to reconstruct the data or target.
- Why can Python be slower than C++ or Java, particularly with large Pandas DataFrames?
H19: Two phones after a change of office
Report date: 2021-11-02 · Role: Quantitative Research Intern · Location: Boston → Chicago
The candidate passed a phone for Boston, but Boston had filled its places, so they accepted another phone for Chicago. Two researchers alternated during that second interview, which lasted more than 80 minutes and ended in rejection.
First phone
- Design an object that returns the mean, maximum and median of incoming values. Packages were not allowed; the discussion focused on the data structures.
- What is the probability that the sum of ten die rolls is 2 modulo 6?
Second phone
- Use gradient descent to solve x^n = a.
- Given trading time series for three stocks, predict the average trading price of another correlated stock. The account does not give the fields, horizon or data. The candidate spent a long time understanding the setup and was unsure which information could be used to model it.
H20: Ordering powers of two and three
Report date: 2021-10-19 · Role: Quantitative Researcher; seniority not stated
Introductions and the resume came first, followed by one algorithm and repeated requests to improve it. The whole phone lasted about 30 minutes.
Technical phone
- Order numbers described as powers of 2 and 3, then keep improving the algorithm. The candidate did not spell out whether this meant separate powers or numbers with mixed factors of 2 and 3.
H21: Mathematics followed by an integration calculator
Report date: 2021-10-16 · Role: Quantitative Researcher
The live phone lasted about 45 minutes. Several mathematics questions came before a CodePair task.
Technical phone
- The interviewer began with a simple differentiation question; the expression was not recalled.
- Find the density of the sum of two normal random variables. Their parameters and dependence were not retained.
- A third mathematics question was forgotten.
- On CodePair, write an integration calculator that takes a function and interval. There were implementation follow-ups, but the candidate did not record them.
H22: Taking numerical integration into more dimensions
Report date: 2021-10-11 · Role: Quantitative Research Intern
This was the candidate's first internship interview. It combined mathematics with CodePair, and a later clarification said there was no Zoom call. Rejection arrived four days later.
Technical phone
- For standard-normal X, derive the density of X².
- An integration question followed, without a retained integrand or interval.
- On CodePair, approximate the integral of f from a to b. Discuss optimization and time complexity, extend it to two dimensions and then m dimensions, and explain the exponential growth of a grid-based approach.
H23: Uniform sums, an absorbing walk and random hats
Report date: 2021-10-01 · Role: Quantitative Research Intern · Location: Shanghai
The Shanghai phone lasted about 45 minutes. A brief introduction was followed by three mathematical questions.
Technical phone
- For independent U[0,1] variables, find the probability that their sum is below 1. Start with two and three variables, then generalize.
- An asymmetric random walk starts at 1 and is absorbed at 0. Find the expected time to absorption and give a rigorous proof. The transition probabilities were not recorded.
- Randomly return n hats to their owners. If X counts correct matches, find E[X] and Var(X), using linearity of expectation.
H24: Parity of heads and an increasing random sequence
Report date: 2021-09-27 · Role: Quantitative Research Intern
The candidate described an unsuccessful 40-minute phone with two mathematical questions.
Technical phone
- After N fair coin flips, what is the probability of an even number of heads?
- Keep drawing U[0,1] values until the sequence first decreases. Find the expected length. The candidate used a sum involving n/n! and reported e, but the account leaves unclear whether the final decreasing draw is counted.
H25: An open-ended one-dimensional prediction question
Report date: 2021-09-01 · Role: Quantitative Research Intern
After introductions and project discussion, most of the phone went into the prediction problem. A statistics question was available if time remained, but it was never reached.
Technical phone
- Introduce yourself.
- You have floating-point arrays xs and ys and a new x. How would you predict y? The candidate explicitly clarified that both x and y are one-dimensional. The interviewer left the modeling choice open.
H26: From abc subsequences to ab and back
Report date: 2021-05-04 · Role: Quantitative Research Intern
Coding filled the 30-minute CodePair phone. The additional technical questions mentioned in the invitation were not reached. Rejection arrived two days later.
Technical phone
- Count distinct subsequences matching abc. After an O(n³) brute-force proposal, the interviewer asked for improvement, simplified the target to ab, and then returned to abc. The candidate wrote an O(n) solution for ab. Other technical questions were mentioned but never reached.
H27: Return the subarray, not just its product
Report date: 2021-02-16 · Role: Quantitative Research Intern
A short introduction led into one coding task in a roughly 30-minute phone. While writing the account, the candidate received an invitation to the next stage.
Technical phone
- Find and return the contiguous subarray with the greatest product. The candidate distinguished returning the subarray from returning only its product. They did not personally receive the negative-number follow-up discussed by other candidates.
H28: A fixed gradient-descent step meets a large input
Report date: 2021-02-16 · Role: Quantitative Researcher; seniority not stated
The virtual onsite had coding in the first and third rounds and statistics/ML in the second.
Final interview 1
- The first final-round interview was mainly coding; the task was not retained.
Final interview 2
- The next covered statistics and ML, without individual questions in the account.
Final interview 3
- Find the positive root of x^N = a, with a > 0 and integer N. The candidate first suggested Newton's method, then was told to use gradient descent. For N=2, a=9, an early result was about 2.99. Other ordinary cases worked; a very large a caused overflow or divergence with a fixed 0.1 step. Smaller step sizes kept being challenged. The candidate treated
g(x)=x^N−aas a derivative and choseF(x)=x^(N+1)/(N+1)−a*x+C, withF′′(x)=N*x^(N−1)>0for positive x. They explained this choice but were unsure whether it was what the interviewer wanted.
H29: Maximum products with fractions, zeros and negatives
Report date: 2021-02-15 · Role: Quantitative Research Intern
The first internship phone lasted about 30 minutes and moved directly from introductions to coding. There was no time to implement the final negative-number follow-up, so the candidate explained it orally.
Technical phone
- For positive values, find the contiguous subarray with the greatest product and return its start and end. Values may be below 1, and an empty choice with product 1 was allowed. Follow-ups introduced zeros and then negative values; the last case was discussed rather than coded.
H30: A fractional-value maximum-product example
Report date: 2021-01-20 · Role: Quantitative Research Intern
The 30-minute phone consisted of introductions and one coding question.
Technical phone
- Return the contiguous subarray of positive numbers with the greatest product. The reported example was [0.1,4,0.8,10,0.1,0.2], giving [4,0.8,10].
H31: Two final rounds, including Pandas and traversal
Report date: 2021-01-06 · Role: Quantitative Research Intern
The candidate completed two final interviews that day within a 75-minute invitation. The first covered finance, mathematics, traversal and Pandas; the second returned to a resume project.
Final interview 1
- The candidate volunteered a finance background and received finance follow-ups, without exact questions in the account.
- Two straightforward mathematics questions were also asked, but not recalled.
- Which is faster to traverse: an array or a linked list?
- Implement some basic Pandas operations. Looking things up was allowed, but the account does not identify the operations.
Final interview 2
- The second final-round interview returned to resume projects.
Keep practicing
- From these reports: search rotated sorted array (H15), timestamp window maximum (H18), positive nth root gradient descent (H18), Why does Ridge help? (H18), stream median mean max (H19), order powers of two and three (H20), predict y (H25), maximum product subarray (H27).
- 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.