Jane Street Interview Questions — Page 1

Jane Street·Software Engineer·Technical Phone Screen

Apr 2026

Jane Street SWE interview with a Connect Four variant problem that had a twist I hadn't seen before. The design scope was pretty wide and they wanted complexity analysis, edge case handling, and actual code, all in one shot.

Design and implement a Connect Four variant where dropping a disc into a column pushes existing discs upward instead of stacking downward, with ejection when the column overflows. Support configurable board size, win detection in all directions, and full undo/redo functionality. Analyze time and space complexity and handle edge cases like invalid columns, ejected discs, turn alternation, and draw detection.

“The push-up mechanic threw me at first.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

Apr 2026

Jane Street data scientist interview with a single deep-dive ML design question around DNA sequence classification. The problem was layered enough that I kept second-guessing whether they wanted a practical answer or a research-level one. Walked away unsure if I hit the right depth.

You have a small labeled dataset of real and fake DNA sequences, plus a much larger set of confirmed real sequences only. Design an end-to-end ML system to classify whether a new DNA sequence is real or fake, covering representation, problem framing, synthetic negatives, model choice, class imbalance, evaluation, threshold calibration, and failure modes specific to biological sequences.

“This one sprawled in every direction.”

View Post

Jane Street·Machine Learning Engineer·Technical Phone Screen

Apr 2026

Jane Street ML Engineer interview, got hit with a pretty involved marketplace simulation problem split into two parts. The second part especially had a lot of moving pieces and I kept second-guessing my pool management logic. Not the kind of question you can brute-force your way through.

Design and implement a DataStore for a marketplace, then write a function that processes buy orders against it: each order fills against the cheapest qualifying unsold listing at or below the buyer's limit price, with ties broken by insertion order. Return a map of filled orders to their final transaction prices.
Implement an end-of-day reconciliation system that processes a transaction log in timestamp order, maintains a pool of re-opened units from failed payments, and produces billing adjustments (refunds or charges) when cheaper units become available for successful buyers or when out-of-stock buyers can now be served from the pool.

“The DataStore part was fine, basically a wrapper with insert and getAll.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

Apr 2026

Went through a probability and combinatorics problem set for a Data Scientist role at Jane Street. The whole interview was built around one coin-flipping game with three parts that escalated in difficulty, plus a handful of follow-ups. Felt like a math exam more than a job interview.

Two players flip a fair coin. Player 1 scores on heads, Player 2 on tails. The first to accumulate 4 of their side wins and the game stops. What is the maximum number of rounds the game can possibly last before a winner is guaranteed?
What is the probability that the game actually requires all 7 rounds to determine a winner, meaning the 7th flip is the deciding one?
Given that the game lasted all 7 rounds, and given that among the first three rounds there were exactly 2 heads and 1 tail, what is the probability that Player 1 won?
Generalize the game: if Player 1 needs n heads and Player 2 needs n tails, what is the maximum game length and what is the probability the game goes to the final possible round?
If the coin is biased with P(heads) = p, and you know the game lasted 7 rounds, what is the probability Player 1 won? Does knowing the composition of the first three rounds matter in this case?
What is the expected number of rounds a fair-coin version of this game lasts?

“This part felt easy and I almost rushed past it.”

View Post

Jane Street·Frontend Engineer·Technical Phone Screen

Apr 2026

Jane Street frontend interview with a data structure design problem that felt more like a systems/algorithms question than anything UI-related. The problem was well-scoped but the performance angle made it tricky.

Design a class that manages a binary field layout: support initializing from a list of named fields with sizes, querying the byte offset of any field by name, and inserting a new field at an arbitrary position in the field order such that subsequent fields shift accordingly.

“My first instinct was just a plain array of (name, size) pairs and compute offsets on the fly.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

Apr 2026

Jane Street Data Scientist interview with a multi-part probability puzzle built around coin tosses and a fictional 'magic wand' mechanic. The problem escalates nicely across three stages, and the follow-ups get genuinely hard. Felt more like a math competition than a typical DS screen.

Four fair coins are tossed simultaneously and you're paid $1 per head. What is your expected payoff?
You're given a wand that lets you flip any two coins of your choice after seeing the initial toss, as many times as you want for free. What's a fair price for this wand?
Now the wand re-tosses any two chosen coins rather than flipping them deterministically. Same rules: free, unlimited uses, you pick when to stop. What's the fair price for this version?
How do the two wand prices behave as you generalize to n coins instead of four?
If each use of the re-toss wand costs a fee c, how does your strategy change and when is the wand worth buying at all?
What if the re-toss wand could only be used at most k times? Set up the dynamic program for its value without fully solving it.

“Straightforward linearity of expectation.”

View Post

Jane Street·Software Engineer·Technical Phone Screen

Apr 2026

Jane Street SWE interview with a heavy ML/systems flavor. The whole thing was built around a CNN evaluation notebook for FashionMNIST, and each part layered on more complexity. Felt more like a research problem set than a traditional coding screen.

You're given a trained CNN and a test set. Implement a row-wise reveal evaluation where at step k, the top k rows are visible and the rest are replaced by a fixed mask value. Record predictions at each k, compute accuracy vs k across the test set, plot the curve, and explain what it tells you about how much information the model actually needs.
Define a reward function for partial image reveals where the reward equals the number of still-masked pixels if the final prediction is correct, and zero otherwise. Using your accuracy-vs-k results, propose and implement a method to find a single global mask fill value that maximizes expected reward across the dataset. Also discuss trade-offs like class imbalance and distribution shift introduced by masking.
Propose a training-time augmentation strategy that masks contiguous row blocks so the model learns to handle partially revealed images. Specify the augmentation policy including probability, region size range, and fill value. Explain how you'd prevent degenerate cases like masking nearly all pixels, and how you'd tune the policy. If you only had two retraining runs, what exact configurations would you try and what metrics would you use to compare them?
With the model fixed, pixels are revealed one at a time at test time. Design an early-exit policy that decides when to stop revealing pixels in order to maximize expected reward. Propose a concrete strategy, describe how to calibrate it offline, and explain how you'd handle ties or oscillations in the model's predictions.

“This was the setup question but it's deceptively meaty.”

View Post

Jane Street·Software Engineer·Technical Phone Screen

May 2026

Jane Street SWE interview with a pretty involved coding problem about a modified connect-k game on an infinite grid. The twist is that dropping a piece pushes the whole column up, which makes win detection way trickier than it sounds.

Two players take turns dropping pieces into columns of an infinite integer grid. Each new piece lands at the bottom of its column and pushes all existing pieces in that column up by one row. Given a required run length k and a sequence of moves, find the first move after which any player has k consecutive pieces in a single row or column, and return which player(s) won. Diagonal lines don't count.

“The column-push mechanic is what got me.”

View Post

Jane Street·Software Engineer·Onsite - System Design / Architecture

May 2026

Jane Street system design round, one big question about designing a file system service from scratch. They wanted the full stack: API design, storage layout, consistency, and how you'd scale it. Pretty intense for a single question.

Design a file system service covering the functional API, data model, storage layout, consistency and concurrency, and scaling strategy. Discuss trade-offs in metadata service design.

“This question is deceptively wide.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

May 2026

Jane Street technical screen for a Data Scientist role. One coding problem, pretty involved, took up most of the time. The problem sounds straightforward until you actually think about the constraints.

You have two event streams, each producing (company, trade_id) pairs. Two streams are considered equivalent if, for every company, the ordered sequence of that company's trade IDs is the same in both streams (interleaving across companies can differ). Write a function that checks equivalence while treating the inputs as streams: single pass, early termination the moment a contradiction is found, and memory bounded by the number of unmatched in-flight events rather than total feed length.

“My first instinct was to group by company, collect all trade IDs per company, then compare.”

View Post