Company interview guide
Akuna Capital Machine Learning Interview Guide (2026)
For ML, AI and research roles: a proctored Python coding test, a camera-on math test, probability interviews, then a final round on statistics, ML and your research.
Akuna Capital's machine learning interview loop has 6 reported stages (Application & role choice, Python coding test, Math & probability test (research track), Video interview & phone screen: probability, Final round: harder probability & coding, Final round: statistics, ML & research); typical length: Several weeks.
Roles covered: Research Scientist, ML Engineer, Data Scientist and Deep Learning Engineer. The question bank holds 23 questions reported from Akuna Capital interviews, 16 of them free to solve in your browser.
- Interview process length
- Several weeks
- Stages
- 6
- Reports linked to bank questions
- 23
- Bank report dates
- Jun 2017–Jul 2026
- Questions in bank
- 23 (16 free)
- Last reviewed
- October 2026
How the Akuna Capital interview loop works
Application (campus applicants pick one top-preference quant or tech role per season), a HackerRank Python coding test, a separate camera-proctored math and probability test for research candidates, a video interview and a phone screen on probability, then a final round of several interviews covering harder probability, coding, statistics, ML and your past work. Level, title and team are set only after the final round, and one ML candidate waited about a month for the decision. The trader track's zero-EV bet-sizing test, recorded wager-game round, market-making competition and mental-math drills are trader-only. Few first-hand ML reports exist, so confirm each round with your recruiter.
The Akuna Capital interview process, stage by stage
01Application & role choice
Online application with a PDF resume; campus and intern applicants pick one top-preference quant or tech role for the season
For entry-level and intern quant research roles, applying marks that role as your top choice and rules you out of other quant and tech roles that season; trader roles stay open. Experienced research roles ask for 3+ years in statistics or ML and a research record. The deep learning research posting wants hands-on RNN, LSTM, transformer or temporal-convolution work on sequential data, PyTorch or TensorFlow, and training at scale. Team (for example strategy research, volatility and pricing, low latency, quant development) is decided after the interviews.
What it tests
- Fit between your statistics, ML and Python record and the specific posting
- Whether you chose the right track: signal research, deep learning on time series, data engineering or development
- Graduation date, grades (3.5+ for campus research roles) and work authorization
How to prepare
- Choose your one quant or tech application carefully; you cannot switch to another quant or tech role later in the season.
- Mirror the posting's language: signal and feature research, reproducible ML workflows, sequence models, data pipelines.
- No finance background is required, so spend prep time on probability, statistics and Python rather than markets.
The firm's postings set the one-role rule, the multiple-challenge warning and team placement after interviews; no ML-role report describes this step.
02Python coding test
HackerRank, Python, fullscreen proctored; research track: 3 algorithm problems where all hidden tests must pass (the 2026-27 cycle ran 3 questions in 120 min with a multiple-monitor check); one 2022 ML engineer test was 60 min of Python, SQL and multiple choice
The first technical gate. Compiled accounts of five 2026 research-track sittings show 15 question slots drawn from only 8 problems: the shortest lexicographically smallest substring with exactly k ones, a binary-string operation-maximizing task, the closest pair after sorting, a consecutive-error counter per server, a grid DP for a drone, minimum swaps to sort by popularity, a delivery order by BFS layers, and a max-sum DP with no two consecutive skips. Software tracks get a C++ or Python low-latency coding test instead.
What it tests
- Easy and medium algorithms that pass every hidden test
- Sliding windows and lexicographic tie-breaks on strings
- Cycle decomposition for minimum swaps
- BFS ordering rules and small grid or linear DPs
- For some ML engineer roles: SQL and statistics multiple choice
How to prepare
- Read ordering rules twice: one delivery problem wants BFS layers sorted by (distance, city id), an older variant wants best-first by smallest id.
- On the no-two-consecutive-skips DP, pad with virtual zeros and allow the last item to be skipped; one candidate lost 4 of 15 hidden tests there.
- Practise in plain Python under a 40-minute-per-problem timer, then check empty input, ties and negative results before submitting.
- If your invitation mentions SQL, revise joins, grouping and window functions too.
Free reported questions for this stage
- K Smallest Substring (Shortest Substring With Exactly K Ones)Coding & AlgorithmsEasySolve
- Maximum Subsequence Sum with No Two Consecutive SkipsCoding & AlgorithmsMediumSolve
- Streaming Statistics: O(1) Max, Mean, and ModeCoding & AlgorithmsMediumSolve
- Basic Calculator IICoding & AlgorithmsMediumSolve
- Maximum Star Sum of a GraphCoding & AlgorithmsMediumSolve
- City With Fewest Reachable Neighbors Under a ThresholdCoding & AlgorithmsMediumSolve
- Build a Binary Tree from Preorder and InorderCoding & AlgorithmsMediumSolve
- Count Palindromic SubstringsCoding & AlgorithmsMediumSolve
- Find the Missing and the Repeated Number via XORCoding & AlgorithmsMediumSolve
- Minimum Days to Release All Updates in OrderCoding & AlgorithmsEasySolve
2 more free questions for this stage are in the app.
Format from compiled accounts of the firm's research coding test (2024-2026); 1 ML engineer report (2022) describes a 60-minute Python, SQL and multiple-choice variant. A posting mentions coding challenges.
03Math & probability test (research track)
EasyHire, camera on, pen and paper; about 8-10 questions: about 5 single-choice calculus and linear-algebra items you must reason through aloud before answering, 3 timed probability puzzles of about 5 min each, then a short behavioral block
A second sitting after the coding test. The linear algebra covers rank of products (Sylvester's bound, rank(AB) versus rank(BA)), rank-nullity, eigenvalues from trace and determinant, and nilpotent or idempotent maps. Calculus covers implicit differentiation, substitution, convergence of improper integrals and differentiating an integral with a variable limit. The probability side asks which statements always hold (Jensen, moments of ratios of normals), variance from a moment generating function, lognormal products and covariance of overlapping coin-pattern indicators, plus quick counting.
What it tests
- Rank, null space and eigenvalue identities
- Derivatives, integrals and convergence tests
- Moments, MGFs, normal and lognormal facts
- Probability puzzles in about 5 minutes each
- Explaining your setup clearly before you answer
How to prepare
- Practise narrating the setup out loud before computing; the reasoning is watched on camera.
- Refresh rank-nullity, Sylvester's inequality, trace and determinant identities and Leibniz's rule.
- For 'which always holds' items, hunt for the one counterexample (heavy tails, non-commuting products) rather than proving every option.
Free reported questions for this stage
- 1000 Blocks Down the StaircaseProbabilityEasySolve
No ML-role report mentions it; compiled accounts of the firm's quant research loop (2024-2026) consistently describe this separate camera-proctored math test.
04Video interview & phone screen: probability
A video interview and a phone screen, each reported at 45-60 min: probability, expected value, combinatorics and puzzles with follow-ups
Compiled accounts describe dice, coin and card classics: expected flips or rolls until a pattern, set up as a small recurrence; card problems such as the most likely position of the first ace or three of a kind; conditional traps like two heads given at least one. Expected value under a stopping rule (keep or reroll, repeat-stop dice), betting to a target and the occasional Markov chain steady state also appear. The firm rewards a fast, defensible estimate you then make exact over a stalled derivation.
What it tests
- First-step conditioning and pattern-waiting recurrences
- Linearity of expectation, indicators and symmetry on cards and dice
- Conditional probability traps
- Expected value under a stopping rule
- Holding up when the interviewer changes one assumption
How to prepare
- Say the recurrence aloud, then commit to a number; give a quick estimate if the algebra stalls.
- Reach for indicators and symmetry before brute-force enumeration.
- State the EV and the decision rule explicitly before optimizing any threshold.
Free reported questions for this stage
- Polya Urn Free KicksProbabilityEasySolve
No ML-role report describes these interviews; compiled accounts of the firm's quant research loop list a video interview and a phone screen of 45-60 min on probability.
05Final round: harder probability & coding
Several interviews in one day, onsite or virtual; harder multi-step probability and EV, and more coding for engineering tracks
Compiled accounts describe the final as the hardest variants of earlier material: multi-step probability, trickier expected value and combinatorics, tied back to a decision. Research candidates are expected to know probability and statistics with some options-pricing intuition; software candidates add C++ or Python coding and options questions. The trader track ends with a live trading final that research and ML candidates do not report.
What it tests
- Multi-step probability and expected value
- Combinatorics with symmetry shortcuts
- Coding with attention to complexity and memory (engineering tracks)
- Basic options intuition: what drives an option's price
How to prepare
- Practise the hard tail of the probability set under a timer, out loud, with a partner adding twists.
- Spend one evening on the firm's own options primer; no finance background is assumed.
- For engineering tracks, be ready to discuss memory layout and latency of the code you write.
No ML-role report covers these final interviews; content comes from compiled accounts of the firm's research and software loops.
06Final round: statistics, ML & research
One or more final-round interviews on statistics, ML and your past work; deep learning and data roles add track-specific depth
The one first-hand account, from an ML data engineering candidate, describes practical questions on machine learning and cloud computing. Research postings center on statistical and ML signals, feature research, portfolio optimization and reproducible ML workflows; the deep learning posting on sequence models for non-stationary, autocorrelated time series and large-scale training. Compiled accounts add estimator questions (Monte Carlo estimates, the serial-number estimation problem) and Markov models; a 2026 research question compared Newton's method with gradient descent.
What it tests
- Overfitting and multiple testing on noisy, non-stationary data
- Leakage and time-ordered validation (walk-forward, purging)
- Regularization, shrinkage and linear-model diagnostics
- Optimization behind your models (Newton versus gradient descent)
- Depth on your own research, or on data pipelines and cloud for engineering roles
How to prepare
- Prepare one project end to end: data, features, validation scheme, what failed and why you trust the result.
- Revise non-stationarity, autocorrelation and look-ahead bias; the deep learning posting names them.
- For engineering-leaning roles, be ready to describe cloud and batch or streaming pipelines you have built.
- Follow up politely if you hear nothing; one candidate waited about a month for a decision.
Free reported questions for this stage
- Newton's Method vs Gradient DescentLinear Algebra & OptimizationMediumSolve
One ML data engineering report (2022) describes practical ML and cloud questions at the final round; the rest comes from the firm's postings and compiled accounts.
Questions and themes reported most often
- Short algorithm problems in timed Python: strings, sorting, BFS, small DP (1 reports, python coding test)
- Statistics multiple choice and SQL beside Python in the ML engineer test (1 reports, python coding test)
- Practical ML: overfitting, leakage and validation on time-ordered data (1 reports, final round: statistics, ml & research)
- Cloud computing and data pipeline questions for ML data roles (1 reports, final round: statistics, ml & research)
- Rank, eigenvalue and calculus facts reasoned aloud on camera (math & probability test (research track))
- Expected flips or rolls until a pattern, set up as a recurrence (video interview & phone screen: probability)
- Card-drawing problems: position of the first ace, hand probabilities (video interview & phone screen: probability)
- Estimators and simulation: Monte Carlo estimates and serial-number estimation (final round: statistics, ml & research)
What Akuna Capital asks about
- Coding (algorithms in Python)28%
- Probability & expected value24%
- Statistics & ML on noisy time series22%
- Calculus & linear algebra10%
- Research discussion & fit10%
- SQL & data pipelines6%
Akuna Capital interview prep checklist
- Solved three medium Python algorithm problems in a timed sitting with every edge case handled
- Can explain rank-nullity, Sylvester's bound and eigenvalues from trace and determinant out loud
- Can solve expected-rolls-until-a-pattern and first-ace problems with a stated recurrence or symmetry argument
- Can describe walk-forward validation, look-ahead bias and multiple-testing corrections for a signal model
- Prepared a deep dive on one research or ML project, including what did not work
- Revised SQL joins and window functions if the role touches data
- Confirmed with the recruiter which tests and rounds apply to your track
Tips
- Ask your recruiter which tests apply: research, ML engineering, data and trader candidates get different screens.
- Make timed Python algorithm sets your first priority; the coding test gates almost every track and every hidden test must pass.
- The research coding test reuses a small pool, so drill string windows, cycle-decomposition swaps, BFS ordering and linear DPs.
- Rehearse speaking math aloud before answering; the math test is camera-proctored and the reasoning counts.
- Drill probability and expected value until first-step conditioning and indicator tricks are automatic.
- Know how to validate models on time-ordered data without leakage, and how multiple testing inflates backtest results.
- For deep learning roles, be ready to compare RNNs, LSTMs, transformers and temporal convolutions on noisy time series.
- Skip market-making games, bet-sizing speed tests and mental-math drills unless you also apply to a trader role.
See the rest of the Akuna Capital loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Akuna Capital loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Akuna Capital interview FAQ
How many interview rounds does Akuna Capital have for ML roles?
Candidates report 6 stages: Application & role choice, Python coding test, Math & probability test (research track), Video interview & phone screen: probability, Final round: harder probability & coding and Final round: statistics, ML & research. Application (campus applicants pick one top-preference quant or tech role per season), a HackerRank Python coding test, a separate camera-proctored math and probability test for research candidates, a video interview and a phone screen on probability, then a final round of several interviews covering harder probability, coding, statistics, ML and your past work. Level, title and team are set only after the final round, and one ML candidate waited about a month for the decision. The trader track's zero-EV bet-sizing test, recorded wager-game round, market-making competition and mental-math drills are trader-only. Few first-hand ML reports exist, so confirm each round with your recruiter.
How long does the Akuna Capital ML interview process take?
Reported timelines run Several weeks.
What does Akuna Capital ask in ML interviews?
By reported question mix: Coding (algorithms in Python) 28%; Probability & expected value 24%; Statistics & ML on noisy time series 22%; Calculus & linear algebra 10%; Research discussion & fit 10%; SQL & data pipelines 6%. The most repeated themes: Short algorithm problems in timed Python: strings, sorting, BFS, small DP; Statistics multiple choice and SQL beside Python in the ML engineer test; Practical ML: overfitting, leakage and validation on time-ordered data.
How should I prepare for the Akuna Capital ML interview?
Ask your recruiter which tests apply: research, ML engineering, data and trader candidates get different screens. Make timed Python algorithm sets your first priority; the coding test gates almost every track and every hidden test must pass. The research coding test reuses a small pool, so drill string windows, cycle-decomposition swaps, BFS ordering and linear DPs.
Are there free Akuna Capital interview questions?
Yes. LeetML has 23 questions reported from Akuna Capital interviews, and 16 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- OptiverFor ML, AI and research roles: a long timed assessment battery, a live betting game with a trader, a probability whiteboard, a P&L-graded take-home and a blunt fit final.
- Susquehanna (SIG)For ML, AI and research roles: a timed probability test, regression-heavy technical calls, a trade-data take-home and a long onsite.
- IMC TradingFor ML, AI and research roles at a global options market maker: a track-split timed test, probability and ML-fundamentals rounds, and a fast superday.
- CitadelFor ML, AI and research roles: the firm's quant research loop, heavy on regression, ML theory on noisy data, probability derivations and classic coding.
- Jane StreetFor ML, AI and research roles: multi-part coding judged on clarity, a hands-on ML round, and probability and statistics on research-leaning loops.
- Hudson River TradingFor ML, AI and research roles: a recycled timed coding test, a probability or programming screen, then a final day of code, systems and statistical depth on very noisy data.