Coding interview questions

Given an array of integers nums and an integer target, return the indices of the two numbers that add up to target. Exactly one solution exists; you may not use the same element twice. Narrate brute force → optimal, then code.

Asked at: Amazon, Google, Meta, Microsoft, Apple — classic FAANG warmup · Difficulty: Easy · Pattern: Hash map · complement lookup

~8 min to rehearse aloud

Warmup that reveals whether you reach for hash maps instinctively instead of nested loops.

• Clarify indices vs values
• Brute → optimal narrative
• Self-pair edge case
• Complexity

• Return indices or values? (indices — LeetCode default)
• Duplicates allowed in the array?
• Negative numbers? (yes — hash map still works)
• Guaranteed one answer, or return empty if none?

1. Brute force: check every pair → O(n²). Say it, then improve.
2. Insight: for value x you need target - x. Remember past values in a map value→index.
3. One pass: for each i, if need in map → return; else store nums[i]→i after the check (avoids self-pair).
4. Complexity: time O(n), space O(n).

def two_sum(nums, target):
    seen = {}  # value -> index
    for i, x in enumerate(nums):
        need = target - x
        if need in seen:
            return [seen[need], i]
        seen[x] = i
    return []

int[] twoSum(int[] nums, int target) {
    Map<Integer, Integer> seen = new HashMap<>();
    for (int i = 0; i < nums.length; i++) {
        int need = target - nums[i];
        if (seen.containsKey(need)) return new int[]{seen.get(need), i};
        seen.put(nums[i], i);
    }
    return new int[]{};
}

• All pairs / duplicates → frequency map or multiset.
• Sorted array → two pointers, O(1) extra space.
• Streaming input → same map, bound memory if needed.

• Return all pairs?
• Sorted array variant?
• Memory-constrained stream?

• Ad targeting join keys
• Deduping request IDs in gateways
• Amazon cart coupon matching patterns

Given a string s, find the length of the longest substring without repeating characters. Example: "abcabcbb" → 3 ("abc").

Asked at: Amazon, Google, Meta, Microsoft — Blind 75 staple · Difficulty: Medium · Pattern: Sliding window · last-seen index

~10 min to rehearse aloud

Tests sliding-window fluency — Meta/Amazon medium staple.

• Window invariant
• Last-seen index correctness
• Empty/all-unique edges

• ASCII / Unicode? (map works either way)
• Empty string → 0
• All unique → n; all same → 1

1. Maintain window [left, right] that is always duplicate-free.
2. Advance right. If s[right] was seen at index ≥ left, set left = last[ch] + 1.
3. Update last[ch] = right; track best = max(best, right-left+1).
4. Time O(n), space O(min(n, alphabet)).

def length_of_longest_substring(s):
    last = {}
    left = best = 0
    for right, ch in enumerate(s):
        if ch in last and last[ch] >= left:
            left = last[ch] + 1
        last[ch] = right
        best = max(best, right - left + 1)
    return best

• Longest with at most k distinct?
• Minimum window covering t?

• Session uniqueness checks
• Log tokenization windows
• Rate windows in analytics

Given intervals[i] = [start_i, end_i], merge all overlapping intervals and return the covering non-overlapping set.

Asked at: Amazon, Google, Meta, Microsoft — scheduling / calendar rounds · Difficulty: Medium · Pattern: Sort + linear merge

~10 min to rehearse aloud

Scheduling/calendar signal — sort + linear merge is the expected pattern.

• Sort justification
• Overlap definition
• Touching intervals

1. Sort by start time — required for a single pass.
2. Keep a "current" interval. If next.start ≤ current.end, current.end = max(ends). Else push current and start new.
3. Touching intervals [1,2][2,3]: ask if they merge (usually yes with ≤).
4. Time O(n log n), space O(n).

def merge(intervals):
    intervals.sort(key=lambda x: x[0])
    out = [intervals[0][:]]
    for start, end in intervals[1:]:
        if start <= out[-1][1]:
            out[-1][1] = max(out[-1][1], end)
        else:
            out.append([start, end])
    return out

• Insert Interval into an already-merged list (O(n), no full resort).
• Meeting Rooms II → min heap of end times / sweep line.
• Min removals to make non-overlapping → greedy by end.

• Insert interval?
• Min rooms?
• Min removals?

• Google Calendar free/busy
• AWS capacity reservation windows
• Ad flight dates

Design LRUCache with get(key) and put(key, value) in O(1) average time. Evict the least recently used key when over capacity.

Asked at: Amazon, Google, Meta, Apple, Microsoft — highest cross-company design+code · Difficulty: Medium · Pattern: Hash map + doubly linked list

~15 min to rehearse aloud

Highest cross-company design+code question — composition under O(1) constraints.

• Why map+DLL
• Draw structure
• Update vs insert
• Capacity-1

• Capacity ≥ 1?
• get miss → -1
• put on existing key updates value AND recency
• Thread safety? (usually out of scope unless asked)

1. Why both structures? Map → O(1) lookup. DLL → O(1) reorder / evict if you already have the node pointer.
2. Sentinel head/tail simplify edge inserts/removes.
3. get hit: unlink node, insert after head (MRU), return value.
4. put: if key exists, remove old node; insert new at head; if size > capacity, remove tail.prev and delete from map.
5. Draw the list on the whiteboard before coding helpers.

class Node:
    __slots__ = ("key", "val", "prev", "next")
    def __init__(self, key=0, val=0):
        self.key, self.val = key, val
        self.prev = self.next = None

class LRUCache:
    def __init__(self, capacity):
        self.cap, self.map = capacity, {}
        self.head, self.tail = Node(), Node()
        self.head.next, self.tail.prev = self.tail, self.head

    def _remove(self, node):
        node.prev.next = node.next
        node.next.prev = node.prev

    def _add_front(self, node):
        node.next, node.prev = self.head.next, self.head
        self.head.next.prev = node
        self.head.next = node

    def get(self, key):
        if key not in self.map:
            return -1
        node = self.map[key]
        self._remove(node); self._add_front(node)
        return node.val

    def put(self, key, value):
        if key in self.map:
            self._remove(self.map[key])
        node = Node(key, value)
        self.map[key] = node
        self._add_front(node)
        if len(self.map) > self.cap:
            lru = self.tail.prev
            self._remove(lru)
            del self.map[lru.key]

• LFU?
• TTL?
• Concurrent access?

• Redis approximate LRU
• CDN edge caches
• CPU page cache intuition

Given an m×n grid of '1' (land) and '0' (water), return the number of islands. Land connects 4-directionally (not diagonally).

Asked at: Amazon, Google, Meta — grid / flood-fill classic · Difficulty: Medium · Pattern: DFS / BFS on grid

~10 min to rehearse aloud

Grid DFS/BFS literacy — Amazon/Google classic.

• Component counting
• Visited discipline
• 4 vs 8 connectivity

1. Scan every cell. On unvisited '1', increment count.
2. Flood-fill (DFS or BFS) to mark the whole component visited (flip to '0' or use a visited set).
3. Never revisit. Prefer BFS if recursion depth worries them.
4. Time O(m·n), space O(m·n) worst case.

def num_islands(grid):
    if not grid:
        return 0
    rows, cols = len(grid), len(grid[0])

    def dfs(r, c):
        if r < 0 or c < 0 or r >= rows or c >= cols or grid[r][c] != "1":
            return
        grid[r][c] = "0"
        for dr, dc in ((1, 0), (-1, 0), (0, 1), (0, -1)):
            dfs(r + dr, c + dc)

    count = 0
    for r in range(rows):
        for c in range(cols):
            if grid[r][c] == "1":
                count += 1
                dfs(r, c)
    return count

• Max area of island
• Number of closed islands
• Pacific Atlantic water flow (multi-source DFS)
• Surrounded regions

• Max area?
• Closed islands?
• Pacific Atlantic?

• Map region labeling
• Image connected components
• Game fog-of-war floods

numCourses labeled 0..n-1. prerequisites[i]=[a,b] means take b before a. Return true iff you can finish all courses (DAG / no cycle).

Asked at: Amazon, Google, Meta, Intuit — topological sort / cycle detection · Difficulty: Medium · Pattern: Graph · Kahn BFS or DFS colors

~12 min to rehearse aloud

Cycle detection / topo sort — dependency systems.

• Graph model
• Kahn vs DFS colors
• Order vs boolean

1. Model: directed edge b→a (b unlocks a). Finish iff DAG.
2. Kahn: compute indegrees; queue all indegree 0; pop and decrement neighbors; count processed nodes.
3. If processed == numCourses → true; else cycle → false.
4. DFS alternative: 0/1/2 colors; back-edge to "in stack" = cycle.

from collections import deque, defaultdict

def can_finish(num_courses, prerequisites):
    graph = defaultdict(list)
    indeg = [0] * num_courses
    for a, b in prerequisites:
        graph[b].append(a)
        indeg[a] += 1
    q = deque([i for i in range(num_courses) if indeg[i] == 0])
    taken = 0
    while q:
        cur = q.popleft()
        taken += 1
        for nxt in graph[cur]:
            indeg[nxt] -= 1
            if indeg[nxt] == 0:
                q.append(nxt)
    return taken == num_courses

• Course Schedule II → return any valid order (Kahn visit order).
• Parallel semesters → longest path in DAG / level BFS.

• Course Schedule II?
• Minimum semesters?

• CI pipeline deps
• Package managers
• Airflow/DAG schedulers

coins of various denominations, total amount. Return fewest coins to make amount, or -1 if impossible. Unlimited supply of each coin.

Asked at: Amazon, Meta, Google — unbounded knapsack DP · Difficulty: Medium · Pattern: Dynamic programming · bottom-up

~12 min to rehearse aloud

Unbounded knapsack DP — distinguishes DP from greedy.

• State definition
• Why greedy fails
• Bottom-up loops

1. Define dp[x] = fewest coins to make x; dp[0]=0; else ∞.
2. For x from 1..amount: for each coin c≤x: dp[x]=min(dp[x], dp[x-c]+1).
3. Why not greedy? Counterexample coins=[1,3,4], amount=6 → greedy 4+1+1=3 coins, optimal 3+3=2.
4. Time O(amount·|coins|), space O(amount).

def coin_change(coins, amount):
    INF = amount + 1
    dp = [0] + [INF] * amount
    for x in range(1, amount + 1):
        for c in coins:
            if c <= x:
                dp[x] = min(dp[x], dp[x - c] + 1)
    return dp[amount] if dp[amount] != INF else -1

Coin Change II counts combinations — different transition order/meaning. Do not confuse them in the interview.

• Number of combinations?
• Limited coin counts?

• Change-making POS
• Resource allocation DP
• Game economy crafting

beginWord → endWord, changing one letter at a time; each intermediate in wordList. Return length of shortest transformation sequence (words in path), or 0 if impossible.

Asked at: Google, LinkedIn, Amazon — BFS shortest path in word graph · Difficulty: Hard · Pattern: BFS · implicit graph

~14 min to rehearse aloud

BFS shortest path in an implicit graph — Google classic.

• Model as graph
• Neighbor generation
• Visited discipline

1. Each word is a node; edge if Hamming distance 1.
2. BFS from beginWord; distance = words in path so far.
3. Neighbor gen: for each position try a–z; check set membership.
4. Remove word when enqueued to avoid revisits.
5. If endWord not in wordList → 0 immediately.
6. Time O(N · L · 26); bidirectional BFS is a strong follow-up.

from collections import deque

def ladder_length(begin_word, end_word, word_list):
    words = set(word_list)
    if end_word not in words:
        return 0
    q = deque([(begin_word, 1)])
    while q:
        word, dist = q.popleft()
        if word == end_word:
            return dist
        for i in range(len(word)):
            for ch in "abcdefghijklmnopqrstuvwxyz":
                nxt = word[:i] + ch + word[i + 1 :]
                if nxt in words:
                    words.remove(nxt)
                    q.append((nxt, dist + 1))
    return 0

• Return the path?
• Bidirectional BFS?

• Typo correction graphs
• Chemical edit distance search
• Knowledge graph hops

Design serialize(root)→string and deserialize(string)→tree. Format is your choice; the pair must be invertible.

Asked at: Meta (Facebook), Amazon, Microsoft — tree encoding classic · Difficulty: Hard · Pattern: BFS / preorder with null markers

~14 min to rehearse aloud

Encoding/decoding + tree fluency — Meta favorite.

• Invertible format
• Null markers
• Empty tree

1. Pick BFS level-order with explicit '#' nulls (interview-friendly).
2. Serialize: queue; append values / '#'; join with commas.
3. Deserialize: rebuild root from first token; for each node consume next two tokens as left/right children.
4. Handle empty tree and single-node trees explicitly.
5. Preorder+nulls also works — mention both.

from collections import deque

class Codec:
    def serialize(self, root):
        if not root:
            return ""
        q, out = deque([root]), []
        while q:
            node = q.popleft()
            if node:
                out.append(str(node.val))
                q.append(node.left)
                q.append(node.right)
            else:
                out.append("#")
        return ",".join(out)

    def deserialize(self, data):
        if not data:
            return None
        vals = data.split(",")
        root = TreeNode(int(vals[0]))
        q = deque([root])
        i = 1
        while q:
            node = q.popleft()
            if vals[i] != "#":
                node.left = TreeNode(int(vals[i]))
                q.append(node.left)
            i += 1
            if vals[i] != "#":
                node.right = TreeNode(int(vals[i]))
                q.append(node.right)
            i += 1
        return root

• BST serialize without nulls?
• Compress?

• Protobuf-like tree payloads
• UI component trees
• AST persistence

n non-negative heights (bar width 1). How much water can the elevation map trap after raining?

Asked at: Amazon, Google, Bloomberg — two pointers / monotonic stack · Difficulty: Hard · Pattern: Two pointers · left/right max

~12 min to rehearse aloud

Hard two-pointer / geometry reasoning under pressure.

• Water formula
• Two-pointer justification
• O(1) space

1. Water at i limited by min(tallest left, tallest right) − height[i].
2. Prefix/suffix max arrays → clear O(n) time / O(n) space version — start here if needed.
3. Two pointers: maintain left_max, right_max; always advance the side with the smaller max (that side's water is fully determined).
4. Time O(n), space O(1). Monotonic stack is another valid approach.

def trap(height):
    if not height:
        return 0
    lo, hi = 0, len(height) - 1
    left_max = right_max = water = 0
    while lo < hi:
        if height[lo] < height[hi]:
            left_max = max(left_max, height[lo])
            water += left_max - height[lo]
            lo += 1
        else:
            right_max = max(right_max, height[hi])
            water += right_max - height[hi]
            hi -= 1
    return water

• Return trapped indices?
• Histogram largest rectangle?

• Hydrology sims
• Capacity planning metaphors
• Image pooling analogies

prices[i] is the stock price on day i. Choose one day to buy and a later day to sell to maximize profit. Return the max profit (0 if no profit).

Asked at: Every FAANG — LC 121; top-5 frequency in 2025–2026 reports · Difficulty: Easy · Pattern: One pass · track min price

~6 min to rehearse aloud

One-pass scan with running state — easiest DP gateway.

• Single transaction constraint
• All decreasing → 0

• One transaction only (buy once, sell once).
• Must sell after buy.
• Empty / length-1 → 0.

1. Brute: try every buy/sell pair O(n²) — reject it.
2. Keep min_price seen so far while scanning left→right.
3. At each price, candidate = price − min_price; track best.
4. Time O(n), space O(1).

def max_profit(prices):
    min_price, best = float("inf"), 0
    for p in prices:
        min_price = min(min_price, p)
        best = max(best, p - min_price)
    return best

• LC 122 unlimited transactions → sum all uphill segments.
• LC 123 at most 2 → DP states.
• Cooldown with cooldown → state machine DP (Amazon favorite).

• Unlimited transactions?
• Cooldown + cooldown?

• Simple PnL calculators
• Promo best-discount windows

Given strings s and t, return the smallest substring of s that covers every character in t (including duplicates). Return "" if impossible.

Asked at: Meta, Amazon, LinkedIn — LC 76; #4 in many 2026 frequency lists · Difficulty: Hard · Pattern: Sliding window · need/have counts

~15 min to rehearse aloud

Hard window with counts — Meta speed round staple.

• need/have counters
• Minimal window shrink
• Duplicates in t

1. Build need counts for t; need_unique = number of distinct chars.
2. Expand right; update have counts; when a char's have hits need, increment formed.
3. While formed == need_unique, shrink left; record best window.
4. Time O(|s| + |t|), space O(alphabet).

from collections import Counter

def min_window(s, t):
    need = Counter(t)
    missing = len(need)
    have = {}
    best_len, best = float("inf"), ""
    left = 0
    for right, ch in enumerate(s):
        have[ch] = have.get(ch, 0) + 1
        if ch in need and have[ch] == need[ch]:
            missing -= 1
        while missing == 0:
            if right - left + 1 < best_len:
                best_len = right - left + 1
                best = s[left : right + 1]
            left_ch = s[left]
            have[left_ch] -= 1
            if left_ch in need and have[left_ch] < need[left_ch]:
                missing += 1
            left += 1
    return best

• Permutation in string?
• Find all anagram starts?

• Log field extractors
• DNA motif covering
• Search snippet covering queries

Find the contiguous subarray with the largest sum and return that sum.

Asked at: Amazon, Google, Microsoft — LC 53; Amazon OA staple · Difficulty: Medium · Pattern: Kadane · running best ending here

~8 min to rehearse aloud

Kadane — Amazon OA classic for array DP.

• Extend vs restart
• All-negative arrays

1. At each index: either extend previous run or start fresh at nums[i].
2. best_ending = max(nums[i], best_ending + nums[i]).
3. Track global max. Handles all-negative by picking the largest element.
4. Time O(n), space O(1).

def max_sub_array(nums):
    best = cur = nums[0]
    for x in nums[1:]:
        cur = max(x, cur + x)
        best = max(best, cur)
    return best

• Return the actual subarray indices.
• Circular maximum subarray.
• 2D Kadane (maximal rectangle sum) — Google follow-up.

• Circular max?
• Return subarray bounds?

• Max streak metrics
• Signal processing windows

nums was sorted ascending then rotated at an unknown pivot. Search for target in O(log n). Distinct values.

Asked at: Meta, Amazon, LinkedIn, Microsoft — LC 33 · Difficulty: Medium · Pattern: Binary search · identify sorted half

~12 min to rehearse aloud

Binary search with a twist — Meta/Amazon favorite.

• Identify sorted half
• Invariant
• Duplicates follow-up

1. Standard binary search frame. Mid always sits in a half that is sorted.
2. If nums[lo] ≤ nums[mid]: left half sorted. If target in [lo, mid), search left; else right.
3. Else right half sorted — symmetric check.
4. Draw an example like [4,5,6,7,0,1,2] every time.

def search(nums, target):
    lo, hi = 0, len(nums) - 1
    while lo <= hi:
        mid = (lo + hi) // 2
        if nums[mid] == target:
            return mid
        if nums[lo] <= nums[mid]:
            if nums[lo] <= target < nums[mid]:
                hi = mid - 1
            else:
                lo = mid + 1
        else:
            if nums[mid] < target <= nums[hi]:
                lo = mid + 1
            else:
                hi = mid - 1
    return -1

• Find minimum?
• Duplicates allowed?

• Rotated ring buffers
• Time-wrapped schedules

Given an integer array, return the k most frequent elements. Order of the answer does not matter.

Asked at: Amazon, Google, Meta, Microsoft — LC 347; universal heap question · Difficulty: Medium · Pattern: Hash map + heap / bucket sort

~10 min to rehearse aloud

Heap vs bucket tradeoff — universal medium.

• Count then select
• O(n log k) vs O(n)
• Stability not required

1. Count frequencies in a map O(n).
2. Option A: min-heap of size k → O(n log k).
3. Option B (often preferred): bucket sort by frequency → O(n).
4. Say both; implement one cleanly.

from collections import Counter

def top_k_frequent(nums, k):
    freq = Counter(nums)
    buckets = [[] for _ in range(len(nums) + 1)]
    for val, c in freq.items():
        buckets[c].append(val)
    out = []
    for c in range(len(buckets) - 1, 0, -1):
        for val in buckets[c]:
            out.append(val)
            if len(out) == k:
                return out
    return out

• Top-k by other metrics?
• Approximate top-k?

• Trending topics
• Error-code dashboards
• Search query popularity

Given meeting time intervals [start, end), find the minimum number of conference rooms required.

Asked at: Meta (Facebook), Amazon, Bloomberg — premium classic · Difficulty: Medium · Pattern: Min-heap of end times / sweep line

~10 min to rehearse aloud

Interval + heap — Meta premium classic.

• Sort by start
• Reuse rule
• Peak rooms

1. Sort meetings by start time.
2. Min-heap stores end times of rooms in use.
3. If next start ≥ earliest end, reuse (pop); else allocate (push).
4. Answer = max heap size during the scan (or final size if you track peak).

import heapq

def min_meeting_rooms(intervals):
    if not intervals:
        return 0
    intervals.sort(key=lambda x: x[0])
    heap = []  # end times
    for start, end in intervals:
        if heap and start >= heap[0]:
            heapq.heappop(heap)
        heapq.heappush(heap, end)
    return len(heap)

Merge Intervals / Insert Interval / Non-overlapping — same family. Sweep line with +1 at start and −1 at end also works.

• Merge intervals link?
• Max concurrent online?

• Meeting room products
• Cloud VM concurrent capacity
• Call-center staffing

Given a string containing just '()[]{}', determine if the input string is valid: open brackets closed by the same type in the correct order.

Asked at: Amazon, Google, Meta, Bloomberg — LC 20; common warmup / phone screen · Difficulty: Easy · Pattern: Stack

~6 min to rehearse aloud

Stack literacy warmup across companies.

• Push/pop matching
• Empty stack rules

1. Scan left→right. Push opening brackets.
2. On closing: stack must be non-empty and top must match.
3. End with empty stack.
4. Time O(n), space O(n).

def is_valid(s):
    pairs = {")": "(", "]": "[", "}": "{"}
    stack = []
    for ch in s:
        if ch in pairs.values():
            stack.append(ch)
        elif ch in pairs:
            if not stack or stack[-1] != pairs[ch]:
                return False
            stack.pop()
        else:
            return False
    return not stack

• Longest valid parentheses (Hard).
• Minimum remove to make valid (Meta).
• Generate parentheses (backtracking).

• Longest valid?
• Min add/remove?

• IDE bracket matchers
• Config/JSON validators
• Template engines

Grid of 0 (empty), 1 (fresh), 2 (rotten). Each minute, any fresh orange 4-adjacent to a rotten one becomes rotten. Return minutes until all fresh are rotten, or -1 if impossible.

Asked at: Amazon, Microsoft, Google — LC 994; multi-source BFS favorite · Difficulty: Medium · Pattern: Multi-source BFS on grid

~12 min to rehearse aloud

Multi-source BFS — Amazon graph/grid favorite.

• Queue all sources
• Level = time
• Impossible case

1. Enqueue all initially rotten cells (multi-source BFS).
2. Count fresh oranges.
3. BFS level-by-level; each level = 1 minute; rot neighbors.
4. If fresh remains → -1; else minutes (careful: last wave may add a minute — track correctly).

from collections import deque

def oranges_rotting(grid):
    rows, cols = len(grid), len(grid[0])
    q, fresh = deque(), 0
    for r in range(rows):
        for c in range(cols):
            if grid[r][c] == 2:
                q.append((r, c))
            elif grid[r][c] == 1:
                fresh += 1
    minutes = 0
    while q and fresh:
        for _ in range(len(q)):
            r, c = q.popleft()
            for dr, dc in ((1, 0), (-1, 0), (0, 1), (0, -1)):
                nr, nc = r + dr, c + dc
                if 0 <= nr < rows and 0 <= nc < cols and grid[nr][nc] == 1:
                    grid[nr][nc] = 2
                    fresh -= 1
                    q.append((nr, nc))
        minutes += 1
    return minutes if fresh == 0 else -1

• Walls and gates?
• Shortest path in grid?

• Content freshness propagation
• Infection/simulation jobs
• Warehouse spill models

You are given a list of words sorted lexicographically in an alien language. Derive any valid order of unique letters. Return "" if invalid.

Asked at: Meta, Google — LC 269; hard topo-sort phone-screen closer · Difficulty: Hard · Pattern: Graph · topological sort from sorted words

~15 min to rehearse aloud

Build graph from constraints — Meta/Google hard closer.

• Edge from consecutive words
• Invalid prefix case
• Cycle → empty

1. Compare consecutive words; first differing chars give an edge earlier→later.
2. Invalid if word A is prefix of longer preceding word ("abc" before "ab").
3. Build graph + indegrees; Kahn BFS for a valid order.
4. If cycle / not all letters processed → "".

from collections import defaultdict, deque

def alien_order(words):
    graph = defaultdict(set)
    indeg = {c: 0 for w in words for c in w}
    for w1, w2 in zip(words, words[1:]):
        if w1.startswith(w2) and w1 != w2 and len(w1) > len(w2):
            return ""
        for a, b in zip(w1, w2):
            if a != b:
                if b not in graph[a]:
                    graph[a].add(b)
                    indeg[b] += 1
                break
    q = deque([c for c, d in indeg.items() if d == 0])
    order = []
    while q:
        c = q.popleft()
        order.append(c)
        for nxt in graph[c]:
            indeg[nxt] -= 1
            if indeg[nxt] == 0:
                q.append(nxt)
    return "".join(order) if len(order) == len(indeg) else ""

• Multiple valid orders?
• Verify order against words?

• Locale collation debugging
• Build order from logs
• Schema evolution ordering

Find the kth largest element in an unsorted array. Note it is the kth largest in sorted order, not the kth distinct.

Asked at: Amazon, Meta, Google, Microsoft — LC 215 · Difficulty: Medium · Pattern: Min-heap of size k / Quickselect

~8 min to rehearse aloud

Selection algorithms — heap vs quickselect signal.

• kth largest vs smallest
• Heap size k
• Average O(n) option

1. Min-heap of size k: push all; pop when size > k; peek is answer O(n log k).
2. Quickselect (Hoare): average O(n) — mention for strong signal.
3. Sorting is O(n log n) — acceptable start, then optimize.

import heapq

def find_kth_largest(nums, k):
    heap = []
    for x in nums:
        heapq.heappush(heap, x)
        if len(heap) > k:
            heapq.heappop(heap)
    return heap[0]

K Closest Points to Origin (LC 973) — same heap pattern with distance.

• K closest points?
• Running median?

• Latency percentile approx
• Leaderboard cutoffs
• Priority aging