DRW Coding Interview Questions
25 DRW coding interview problems with full optimal solutions — 8 easy, 12 medium, 5 hard. Every problem ships with multiple approaches (brute-force first, then the optimal), complexity tables for each, company-specific tips on what an DRW interviewer values, and a FAQ section.
Showing 12 problems of 25
- #3mediumvery frequently asked
3. Longest Substring Without Repeating Characters
DRW uses this problem to test the sliding-window pattern — the same technique that powers rolling-window deduplication on market-data feeds and tick-data deduplication in low-latency pipelines. Getting to O(n) with a single hash-map pass is required; naive O(n²) solutions are rejected on sight.
- #15mediumfrequently asked
15. 3Sum
DRW asks 3Sum to test whether candidates can combine sorting with two-pointer sweeps and suppress duplicates without extra memory. In trading contexts this pattern appears in three-asset arbitrage detection: find all triplets of prices summing to zero profit after costs.
- #49mediumfrequently asked
49. Group Anagrams
DRW uses Group Anagrams to test hash-key design — the skill of choosing a canonical representation that correctly groups equivalent items. In trading systems this appears in instrument deduplication: grouping ticker aliases (GOOG / GOOGL, different listings of the same underlying) by a canonical identifier.
- #56mediumfrequently asked
56. Merge Intervals
DRW frames Merge Intervals as an order-book depth aggregation problem: given a set of bid or ask price-quantity intervals, collapse overlapping ranges into consolidated depth levels. Sorting by start time is mandatory; the merge sweep is O(n).
- #139mediumfrequently asked
139. Word Break
DRW surfaces Word Break to test DP on string segmentation — a pattern that maps to parsing FIX tag-value sequences and tokenizing structured market messages where the dictionary of valid tokens is known in advance.
- #146mediumvery frequently asked
146. LRU Cache
DRW uses LRU Cache because low-latency market-data systems rely on exactly this design: a bounded cache of recently-seen instrument snapshots, where the stale entry is evicted on each new quote. O(1) get and put are non-negotiable — the interviewer will ask for the doubly-linked-list proof.
- #200mediumfrequently asked
200. Number of Islands
DRW uses Number of Islands as a graph traversal proxy — the same connected-component logic appears in counterparty exposure clustering and in detecting isolated market segments. BFS and DFS both work; interviewers ask you to compare them on memory use.
- #207mediumfrequently asked
207. Course Schedule
DRW uses Course Schedule to probe topological sort and cycle detection — the same logic used in dependency resolution for trading system component startup ordering and in detecting circular position-dependency chains in risk models.
- #238mediumfrequently asked
238. Product of Array Except Self
DRW uses this problem to test prefix-product reasoning without division — the same technique used to compute portfolio-weight normalization factors and running exposure metrics without re-scanning the full array. The O(n) no-division constraint is the key DRW signal.
- #295mediumvery frequently asked
295. Find Median from Data Stream
Running median over a live stream is a first-class problem at DRW: median bid-ask spread, median fill latency, median position size across strategies. The two-heap technique gives O(log n) insertion and O(1) query. DRW will ask for the Fenwick tree variant when values are bounded integers.
- #322mediumfrequently asked
322. Coin Change
DRW uses Coin Change to test unbounded knapsack DP — and immediately connects it to tick-size discretization: given a set of valid price increments (tick sizes), what is the minimum number of ticks to express an order quantity? The greedy approach fails for arbitrary denominations; DP is required.
- #347mediumfrequently asked
347. Top K Frequent Elements
DRW asks Top K Frequent Elements to probe heap vs. bucket-sort trade-offs — decisions that matter when ranking the k most-traded instruments by volume or the k most-active order IDs in a session. O(n log k) with a heap is expected; O(n) with bucket sort earns extra credit.