Amazon DSA question list (curated)#

Not 500 problems. Amazon repeats a narrow band of patterns plus a few OA-only puzzles. Leadership Principles are in every coding round — finish 5 minutes early for LP.

How this differs from Google

More implementation + hashing + trees + heaps. Fewer “invent a new graph DP.”
OA (online assessment) is its own genre: logs, “most common word,” two-sum variants, grids.
Onsite: 1–2 Mediums, or 1 Medium + 1 Hard. Clean code + tests beat a clever incomplete Hard.
They love design-flavored coding: LRU, autocomplete, hit counter, tic-tac-toe.
Language: Python or Java both fine. Amazon OA often Java/Python/C++.

Volume: ~110–130 unique. [A] = Amazon-recurring. [OA] = online assessment / work-simulation style.

Mark x only after a timed clean re-solve (35 min Medium, 45 min Hard).

Amazon OA — do these first (2 weeks)#

OA is often 2 coding questions, 70–90 min, no human. Optimize for compiling + passing hidden tests.

String / hash (OA staples)#

Arrays / two pointers / greedy#

Sliding window#

3. Longest Substring Without Repeating Characters [A]
209. Minimum Size Subarray Sum
438. Find All Anagrams in a String
76. Minimum Window Substring [A] Hard
239. Sliding Window Maximum [A] Hard
904. Fruit Into Baskets
1004. Max Consecutive Ones III
340. Longest Substring with At Most K Distinct Characters [A] (premium; else 159 / 904)

Heap / top-K / “connect ropes”#

Grid / BFS (OA “zombie / treasure island”)#

Search suggestions / trie OA#

1268. Search Suggestions System [A][OA]
208. Implement Trie [A]
642. Design Search Autocomplete System [A] Hard (premium)

Onsite coding — by pattern#

Linked list#

Trees / BST (Amazon’s densest onsite cluster)#

Graphs (beyond grid)#

Binary search#

Stack / parse / monotonic#

Backtracking#

DP (Amazon asks fewer than Google, but not zero)#

Design-as-code (Amazon onsite loves these)#

Matrix / “read carefully”#

High-signal Amazon uniques (don’t skip)#

These show up in “recent Amazon” reports more than generic NeetCode:

#	Problem	Why
937	Reorder Data in Log Files	OA almost guaranteed historically
819	Most Common Word	OA
1268	Search Suggestions System	OA + trie
1192	Critical Connections	Tarjan bridges — SDE2 tell
1167	Min Cost to Connect Sticks	Huffman / heap
763	Partition Labels	greedy + last index
545	Boundary of Binary Tree	messy tree impl
314	Vertical Order (not 987)	BFS + col index
636	Exclusive Time of Functions	stack of functions
348	Tic-Tac-Toe	O(1) move
146	LRU	every level
200 + 994	Islands / rotting	grid BFS
138	Copy Random List	graph-on-list
297	Serialize tree
253	Meeting Rooms II	heap
973	K Closest	heap / quickselect
692	Top K Words	heap + tie-break
675	Cut Off Trees	multi BFS
472	Concatenated Words	word break on a set
1235	Job Scheduling	DP + bisect
445	Add Two Numbers II	stack, no reverse list follow-up

Target counts#

	Count	Notes
OA cluster	25–30	Do before applying
Medium	80–95	Core
Hard	20–28	LRU internals, islands-hard, Tarjan, serialize, trap rain, merge k
Easy	10	Warmup only
Total unique	~120	Overlap with the Google list is large — don’t double-grind

If you already finished the Google list: add OA strings, Critical Connections, Search Suggestions, Connect Sticks, Partition Labels, Vertical Order 314, Exclusive Time, Tic-Tac-Toe, Add Two Numbers II, Concatenated Words. That delta is the Amazon-specific work.

Live-interview framework (Amazon flavor)#

Clarify (duplicates, sorted, in-place, n=10^5, follow-up constraints).
Example + 1 edge.
Brute → complexity → better. Name the pattern.
Code in the language you’ll use in OA (same as onsite).
Walk an example. Null, empty, 1 element, overflow, disconnected.
Stop with 5 min left. “Tradeoff if this is a stream / concurrent map?”
LP probe: “Tell me about a time you disagreed…” — have 2 stories ready in the same interview.

Amazon rejects working code with silent arrogance and almost-correct code with no tests equally.

Mock sets (45 min, talk aloud)#

Mock	A	B (if two-question)
OA-1	937 Log Files	973 K Closest
OA-2	819 Most Common Word	200 Islands
OA-3	1268 Search Suggestions	763 Partition Labels
OA-4	994 Rotting Oranges	1 Two Sum unique pairs
Onsite-1	146 LRU	56 Merge Intervals
Onsite-2	297 Serialize	215 Kth Largest
Onsite-3	236 LCA	3 Longest Substring
Onsite-4	138 Copy Random	33 Rotated Search
Onsite-5	1192 Critical Connections	20 Valid Parentheses
Onsite-6	23 Merge k	102 Level Order
Onsite-7	239 Sliding Window Max	98 Validate BST
Onsite-8	348 Tic-Tac-Toe	394 Decode String
Onsite-9	721 Accounts Merge	42 Trap Rain
Onsite-10	127 Word Ladder	155 Min Stack

Do OA-1..4 timed without notes before you submit the application.

What Amazon does not need#

Segment trees / Fenwick (unless you already know them)
Max flow
Heavy geometry
Bitmask DP contests
200 Easy
System-design-as-code beyond LRU/LFU/autocomplete/file system

SDE level calibration (coding only)#

Level	Coding bar
SDE 1	Clean Medium in 25–30 min; OA strong; trees + hash + BFS
SDE 2 (L5-ish)	Medium without hints; one Hard (LRU, serialize, trap, merge k, Tarjan); talk tradeoffs
SDE 3	Same plus follow-ups (stream, concurrent, scale the data structure); you lead

Leadership Principles are scored in the coding round, not only in the “bar raiser.” Bias for Action = you pick an approach and code. Dive Deep = you test. Have Backbone = you defend complexity. Earn Trust = you don’t hide a bug.

Python OA kit (if you interview in Python)#

python

from collections import deque, defaultdict, Counter, OrderedDict
import heapq, bisect
# LRU: OrderedDict move_to_end / popitem(last=False)
# logs: digit vs letter — `s.split()[1][0].isdigit()`

Java: ArrayDeque, PriorityQueue, LinkedHashMap(accessOrder=true) for LRU — see java_interview_cheatsheet.md.