Cohere Coding Interview Questions
25 Cohere 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 Cohere interviewer values, and a FAQ section.
- #1easyvery frequently asked
1. Two Sum
Given an integer array and a target, return indices of the two numbers that add up to the target. Cohere asks this as a warm-up because hash-map lookup is the foundation of embedding deduplication and nearest-neighbor retrieval at scale.
- #3mediumvery frequently asked
3. Longest Substring Without Repeating Characters
Find the length of the longest substring with all unique characters. Cohere asks this because sliding-window deduplication is core to tokenization span selection and context-window management in production language-model serving.
- #4hardoccasionally asked
4. Median of Two Sorted Arrays
Find the median of two sorted arrays in O(log(m+n)) time. Cohere asks this hard problem because binary search on partition boundaries is the same mathematical discipline applied to efficient percentile estimation in model evaluation pipelines.
- #15mediumfrequently asked
15. 3Sum
Find all unique triplets in an array that sum to zero. Cohere asks this to test systematic duplicate-elimination alongside the two-pointer technique — skills that transfer directly to deduplicating retrieved document clusters in a RAG pipeline.
- #20easyfrequently asked
20. Valid Parentheses
Determine if a string of brackets is balanced. Cohere uses this as a warm-up to test stack reasoning — the same pattern that validates structured outputs and nested function calls in LLM inference pipelines.
- #21easyfrequently asked
21. Merge Two Sorted Lists
Merge two sorted linked lists into one sorted list. Cohere values this because the merge step is the backbone of k-way merge used when combining ranked candidate lists from multiple retrieval indexes in RAG pipelines.
- #23hardfrequently asked
23. Merge K Sorted Lists
Merge k sorted linked lists into one sorted list. Cohere asks this because k-way merge is the core algorithm for combining ranked results from multiple retrieval shards in a distributed RAG system — minimising latency while maintaining global ordering.
- #42hardfrequently asked
42. Trapping Rain Water
Calculate how much water can be trapped between bars after rain. Cohere asks this because the two-pointer insight — maintaining running maxima from both ends — mirrors how bidirectional attention masks are constructed in transformer architectures.
- #49mediumfrequently asked
49. Group Anagrams
Group strings that are anagrams of one another. Cohere asks this because canonical-key hashing mirrors how embedding models deduplicate semantically-equivalent queries before routing them to a retrieval index.
- #53easyfrequently asked
53. Maximum Subarray
Find the contiguous subarray with the largest sum. Cohere asks this because Kadane's algorithm demonstrates the greedy DP reasoning that underlies attention-window selection and relevance-score aggregation in retrieval pipelines.
- #56mediumfrequently asked
56. Merge Intervals
Merge all overlapping intervals in a list. Cohere asks this because interval merging directly models context-window deduplication, log event collapsing, and bounding-box merging in document-layout analysis for enterprise RAG systems.
- #70easyfrequently asked
70. Climbing Stairs
Count distinct ways to reach the nth stair taking 1 or 2 steps at a time. Cohere uses this to gauge whether candidates recognise overlapping subproblems — a mental model critical for understanding dynamic-programming approaches in sequence modelling.
- #121easyfrequently asked
121. Best Time to Buy and Sell Stock
Find the maximum profit from a single buy-sell transaction. Cohere includes this because the running-minimum pattern mirrors how inference cost optimizers track cheapest-batch windows across fluctuating GPU pricing.
- #127hardoccasionally asked
127. Word Ladder
Find the shortest transformation sequence from one word to another, changing one letter at a time. Cohere asks this because BFS over an implicit graph is the same reasoning used to find shortest edit-distance paths between token sequences in vocabulary-constrained generation.
- #139mediumfrequently asked
139. Word Break
Determine if a string can be segmented into words from a dictionary. Cohere asks this because the DP segmentation pattern directly models how subword tokenisers decide optimal byte-pair-encoding splits during vocabulary construction.
- #146mediumvery frequently asked
146. LRU Cache
Design a cache that evicts the least-recently-used entry when full. Cohere asks this because LRU caching is a first-class concern in embedding serving — caching recently-requested embedding vectors avoids redundant inference and reduces GPU cost at scale.
- #200mediumfrequently asked
200. Number of Islands
Count the number of connected components in a binary grid. Cohere asks this to test graph traversal — the same connected-component reasoning used when clustering retrieved document chunks into coherent topic groups.
- #206easyfrequently asked
206. Reverse Linked List
Reverse a singly-linked list in place. Cohere interviewers use this to probe pointer manipulation — the same discipline required when implementing custom tokenizer chains and streaming decoder buffers.
- #207mediumfrequently asked
207. Course Schedule
Detect whether a set of course prerequisites can all be satisfied — i.e., detect a cycle in a directed graph. Cohere asks this because topological ordering of task dependencies mirrors pipeline DAG scheduling for multi-step RAG and agentic workflows.
- #238mediumfrequently asked
238. Product of Array Except Self
Compute the product of all elements except each element itself, without division. Cohere asks this because the prefix/suffix product pattern is the same technique used in attention-mask generation and cumulative probability normalisation in beam search.
- #283easyoccasionally asked
283. Move Zeroes
Move all zeroes to the end of an array while preserving the relative order of non-zero elements. Cohere asks this to test in-place two-pointer discipline — the same pattern used when compacting sparse token arrays in a batched inference engine.
- #322mediumfrequently asked
322. Coin Change
Find the minimum number of coins that sum to a target amount. Cohere uses this to assess unbounded knapsack DP — the same optimisation framework underlying minimum-edit decoding and optimal token-budget allocation in constrained generation.
- #347mediumfrequently asked
347. Top K Frequent Elements
Return the k most frequent elements in an array. Cohere asks this because frequency-based ranking is central to term-frequency scoring, vocabulary selection during tokeniser training, and log-aggregation pipelines.
- #642hardoccasionally asked
642. Design Search Autocomplete System
Design a real-time autocomplete system that suggests the top-3 most frequently typed sentences matching the current prefix. Cohere asks this design-heavy hard problem because prefix-indexed retrieval with frequency ranking is the core of their enterprise search and query-suggestion products.
- #692mediumfrequently asked
692. Top K Frequent Words
Return the k most frequent words, with lexicographic tiebreaking. Cohere includes this because frequency-ranked vocabulary selection with deterministic tiebreaking is exactly how tokeniser vocabularies are built from a training corpus.