Company interview guide

Five Rings Machine Learning Interview Guide (2026)

For ML, AI and research roles: a camera-on no-calculator math screen, two live probability and statistics rounds, then a research-track data project.

Five Rings's machine learning interview loop has 6 reported stages (Application, Timed mental-math and estimation screen, Technical phone: probability, Technical phone: estimation, inference and regression (research track), Take-home data project (research track), Final rounds); typical length: Several weeks; about a week between rounds is commonly reported.

Roles covered: Research Scientist, Applied Scientist, Data Scientist and ML Engineer. The question bank holds 12 questions reported from Five Rings interviews, 5 of them free to solve in your browser.

Try a free question

Interview process length
Several weeks; about a week between rounds is commonly reported
Stages
6
Reports linked to bank questions
12
Bank report dates
Mar 2021–Oct 2025
Questions in bank
12 (5 free)
Last reviewed
October 2026

How the Five Rings interview loop works

Application, then a timed no-calculator screen (an online test of roughly 12-20 fill-in questions, or a version read aloud by HR), a 45-60 minute probability phone round with a researcher, a 45-60 minute estimation and inference round, a take-home data project under NDA, and final rounds with senior staff. The 30-second phone math, Fermi estimation section, dice and betting design games and head-of-trading round are trader-track only and get no stage here. The software developer track is not covered. The outline rests on compiled accounts, not first-hand ML reports, so read the order and timings as the usual shape rather than a guarantee.

The Five Rings interview process, stage by stage

01Application

Online application: resume, education with GPA, outstanding offers, preferred department and work authorization; research roles are in New York

The firm has offices in New York, Boca Raton, London and Amsterdam; both research postings are in New York. The experienced researcher posting asks for two or more years building quantitative strategies in futures or equities, intraday signal work on large financial datasets for mid- to high-frequency trading, and R or Python. The PhD internship runs nine weeks over the summer with a mentor and asks only for basic programming. Neither posting mentions machine learning or the interview stages.

What it tests

  • A deep quantitative background: math, statistics, physics, computer science or economics
  • Hands-on statistical analysis of large datasets in R or Python
  • For lateral hires, a record of finding intraday signals in futures or equities

How to prepare

  • Lead the resume with statistical results (out-of-sample checks, effect sizes) rather than ML frameworks; the postings never ask for them.
  • Fill in outstanding offers and deadlines; replies are slow and the form invites it.
  • PhD intern applicants need no finance; spend the time on probability speed instead.

The firm's job board and research postings describe the application form and required background; no first-hand ML-role report covers this step.

02Timed mental-math and estimation screen

Proprietary online test, webcam on, no calculator, no coding: about 17-20 fill-in questions in 17-40 minutes, or 12-16 with per-question timers of roughly 20-80 seconds on the research version; sometimes read aloud by HR at about 30 seconds a question

The stage most candidates fail, on every track. Answers are typed numbers with no options, to the nearest integer where an estimate is asked: dice, coin and card probability, expected value of short games, conditional probability and Bayes, counting, modular arithmetic, short sequences, and quick approximations such as 1.01^20, the tenth root of 10, the square root of 228, the digits of 50! or an arc length. The research version skews toward distributions and inference. Accounts mention a forced pause of about ten seconds before each question, no going back, and occasional platform stalls; no negative marking is reported, so always enter a number.

What it tests

  • Mental arithmetic by decomposition with no calculator
  • Instant expectation and conditional probability by counting the sample space
  • Bounding a fast-growing quantity to the nearest integer instead of computing it
  • Recognizing a distribution identity before the algebra gets long
  • Staying accurate under a visible per-question clock

How to prepare

  • Drill two-minute arithmetic sprints daily, plus timed sets of 20 questions in 20 minutes or less.
  • Memorize anchors (ln 2, ln 10, log10 of 2, square roots up to 20) so powers, roots and factorials become one-line estimates.
  • Write the denominator first on probability items; the numerator is usually a short count.
  • If an item needs more than a line of algebra, bound it, answer and move on; a blank scores the same as a wrong number.

Free reported questions for this stage

Format and content from compiled accounts of the firm's research and trading screen, consistent across the 2023-2026 cycles; no first-hand ML-role report.

03Technical phone: probability

45-60 minutes by phone or video with a quant researcher, worked out loud

Deeper probability with a researcher, often a math PhD who gives hints freely. Recurring themes are the variance of one coordinate of a random point on a sphere, P(X+Y > XY) for two dice, the expected length of the longer piece of a broken stick, random-walk hitting probabilities, and the bus-stop waiting time, where length-biased sampling makes the expected wait longer than half the mean gap. Regression and correlation facts can also come up here (see the next stage). Interviewers grade the argument as much as the number.

What it tests

  • Symmetry and indicator arguments that remove casework
  • Geometric probability on sticks, circles and spheres
  • Random walks, first-step analysis and hitting probabilities
  • The inspection paradox and length-biased sampling
  • Conditioning with jointly normal variables

How to prepare

  • Say which symmetry or indicator you will use before computing; it often turns the problem into one line.
  • Know the bus-stop residual-wait result and derive it from the gap distribution.
  • Take hints quickly and say how they change your plan; defending a stuck approach costs more than the hint.
  • Sanity-check each answer at an extreme case before stating it.

Free reported questions for this stage

Format and content from compiled accounts of the firm's quant research first phone round; the problems placed here carry research and trader tags. No ML-role report.

04Technical phone: estimation, inference and regression (research track)

45-60 minutes with a researcher; derivations talked through or on a shared document

The research track's second technical round and the closest the loop gets to ML theory. Reported prompts: the maximum likelihood estimator for n iid draws from U(0, theta), its bias correction and mean squared error against a moment estimator; confidence intervals with unknown variance against a distribution-free interval; the number of U(0,1) draws until the sum exceeds 1 (mean e) with its variance and an interval for a simulated estimate; mean against median for a decreasing density via Jensen; and regression facts: the slopes of Y on X and X on Y, whose product is R-squared, and correlation recovered from the spreads of X, Y and X+Y.

What it tests

  • Likelihood, bias, variance and MSE of competing estimators
  • Normal-theory against distribution-free confidence intervals
  • Regression slopes in both directions and their link to correlation
  • Valid correlation structures, including bounds on equal pairwise correlation
  • Jensen-style comparisons of mean and median

How to prepare

  • Derive the distribution of a sample maximum and its first two moments in a few lines.
  • When asked which estimator you would use, answer with MSE and robustness, not only unbiasedness.
  • Check every correlation answer against the -1 to 1 range and positive semi-definiteness.
  • For a simulation answer, give an interval as well as the point estimate.

Free reported questions for this stage

Format and content from compiled accounts of the firm's research-track second phone round; the four problems placed here carry research tags. No ML-role report.

05Take-home data project (research track)

Take-home analysis of a dataset under NDA; time limit, tools and deliverable are not publicly reported

Research candidates get a dataset to analyze on their own time before the final rounds. The NDA keeps details scarce, but a weak submission is reported as a common reason for rejection, so it is a real gate. Given the research postings' focus on intraday signals in large financial datasets, prepare for time-ordered data where the main risk is a result that does not survive honest out-of-sample testing.

What it tests

  • Clean exploratory analysis and data checks
  • A simple, defensible model with honest validation
  • No look-ahead: time-ordered splits and features built only from past data
  • A clear write-up whose main finding a researcher can check quickly

How to prepare

  • Ask the recruiter for the time limit, allowed tools and expected deliverable before you start.
  • Prefer a well-validated linear or tree baseline to a deep model, and show how the edge holds up period by period out of sample.
  • Count how many variants you tried and say so; a result picked from many attempts needs a stricter bar.
  • Be ready to defend every choice in the final rounds.

Described only in compiled accounts of the firm's quant research loop; the firm's postings do not mention it, and no first-hand report gives its length or format.

06Final rounds

Several conversations with senior researchers and a senior leader; trader candidates see a head-of-trading round first

The closing rounds mix more probability and expected-value reasoning with fit: why markets, why the firm, and how you work with traders and developers. The firm is small, so team fit carries weight, and the full-time bar is reported as very high. Research candidates may be asked about their take-home or past research.

What it tests

  • Expected-value reasoning under follow-up questions
  • Discussion of your take-home or past research
  • Why trading research and why the firm
  • Working with traders and software developers

How to prepare

  • Keep probability practice going until the end; the finals still test it.
  • Prepare a short, specific answer on why market research over a tech or academic ML role.
  • Keep other processes moving; replies are slow.

Compiled accounts of the firm's research and trading loops describe senior final rounds; research-track format and length are not reported first-hand.

Questions and themes reported most often

  • Linearity of expectation with indicators: fixed points of a permutation, components in a random coloring (technical phone: probability)
  • Uniform upper endpoint: MLE, bias correction and MSE against the sample mean (technical phone: estimation, inference and regression (research track))
  • Regression and correlation: slopes both ways, correlation from the spread of a sum, equal pairwise bounds (technical phone: estimation, inference and regression (research track))
  • Geometric probability: broken stick, random triangle on a circle, a coordinate of a point on a sphere (technical phone: probability)
  • Normal variables: conditional expectation given the sum, moments of a ratio (technical phone: probability)
  • Draws of U(0,1) until the sum exceeds 1: mean e, its variance and an interval (technical phone: estimation, inference and regression (research track))
  • Bus-stop waiting time and the inspection paradox (technical phone: probability)
  • Bayes update for a coin's bias after a few heads (timed mental-math and estimation screen)

What Five Rings asks about

  • Probability & expectation30%
  • Statistics, estimation & regression25%
  • Time-series modeling & validation (take-home)20%
  • Mental math & estimation speed15%
  • Research discussion & fit10%

Five Rings interview prep checklist

  • Scored 15 or more out of 20 on a timed no-calculator set in 20 minutes
  • Can estimate 1.01^20, the tenth root of 10 and log2(50!) to the nearest integer in under a minute each
  • Solved fixed points, the broken stick and the sphere-coordinate variance with symmetry or indicators
  • Can derive the bus-stop expected wait from the gap distribution
  • Compared the sample-maximum and sample-mean estimators for U(0, theta) by MSE
  • Can explain why the two regression slopes multiply to R-squared and recover correlation from the spread of X+Y
  • Ran a timed practice analysis on market-style data with walk-forward validation and a one-page write-up
  • Have a clear answer to why trading research over an ML or tech role

Tips

  • Treat the timed screen as the hardest gate: drill it daily with no calculator and a visible clock for at least two weeks.
  • Reach for indicators, symmetry and first-step analysis before casework; most reported problems fall to them.
  • Reason out loud; interviewers hint freely and grade the argument as much as the answer.
  • Know the uniform-endpoint estimators, the two regression slopes and the sum-of-uniforms result cold.
  • Tie closed-form answers to ML practice: bias-variance, attenuation from noisy regressors, autocorrelated samples.
  • On the take-home, validate in time order, keep the model simple and report how many variants you tried.
  • Skip the trader-only dice-design and betting games unless your recruiter says your track includes them.
  • Expect about a week between rounds and keep other applications moving.

See the rest of the Five Rings loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the Five Rings loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Five Rings interview FAQ

How many interview rounds does Five Rings have for ML roles?

Candidates report 6 stages: Application, Timed mental-math and estimation screen, Technical phone: probability, Technical phone: estimation, inference and regression (research track), Take-home data project (research track) and Final rounds. Application, then a timed no-calculator screen (an online test of roughly 12-20 fill-in questions, or a version read aloud by HR), a 45-60 minute probability phone round with a researcher, a 45-60 minute estimation and inference round, a take-home data project under NDA, and final rounds with senior staff. The 30-second phone math, Fermi estimation section, dice and betting design games and head-of-trading round are trader-track only and get no stage here. The software developer track is not covered. The outline rests on compiled accounts, not first-hand ML reports, so read the order and timings as the usual shape rather than a guarantee.

How long does the Five Rings ML interview process take?

Reported timelines run Several weeks; about a week between rounds is commonly reported.

What does Five Rings ask in ML interviews?

By reported question mix: Probability & expectation 30%; Statistics, estimation & regression 25%; Time-series modeling & validation (take-home) 20%; Mental math & estimation speed 15%; Research discussion & fit 10%. The most repeated themes: Linearity of expectation with indicators: fixed points of a permutation, components in a random coloring; Uniform upper endpoint: MLE, bias correction and MSE against the sample mean; Regression and correlation: slopes both ways, correlation from the spread of a sum, equal pairwise bounds.

How should I prepare for the Five Rings ML interview?

Treat the timed screen as the hardest gate: drill it daily with no calculator and a visible clock for at least two weeks. Reach for indicators, symmetry and first-step analysis before casework; most reported problems fall to them. Reason out loud; interviewers hint freely and grade the argument as much as the answer.

Are there free Five Rings interview questions?

Yes. LeetML has 12 questions reported from Five Rings interviews, and 5 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources