Question: Check if string of ()[]{} is valid. "([)]" = false.

Expected Answer: Use Stack. Push opening, pop on closing and check match. O(n) time, O(n) space.

bool IsValid(string s) {
    var stack = new Stack<char>();
    var map = new Dictionary<char,char>{{')','('},{'}','{'},{']','['};
    foreach(char c in s) {
        if("({[".Contains(c)) stack.Push(c);
        else if(stack.Count==0 || stack.Pop()!=map[c]) return false;
    }
    return stack.Count==0;
}

Why they ask: Tests stack basics + edge cases. Empty string = true. One char = false.
Follow-up: What if input has other chars? Ignore them or return false - clarify with interviewer first.

Question: You return 1M users from DB. Return type: IEnumerable<User> vs List<User> vs IQueryable<User>?

Expected Answer:
1. IQueryable<User>: Best for DB. Keeps query composable. db.Users.Where() adds to SQL, not in-memory.
2. IEnumerable<User>: For in-memory or streaming. Deferred execution, but if from EF Core, each iteration hits DB.
3. List<User>: Worst. Forces .ToList(), loads 1M to RAM = OOM. Only use if you need Count or multiple iterations.
Amazon trap: Calling .Count() on IQueryable runs SELECT COUNT(*). Calling it on IEnumerable loads all then counts = slow.

Question: Design LRU Cache with Get(key) and Put(key,val) in O(1).

SDE-1 Answer: Dictionary + Doubly Linked List. Dictionary for O(1) lookup, DLL for O(1) remove/add to head.

public class LRUCache {
    class Node { public int key,val; public Node prev,next; }
    Dictionary<int,Node> map = new(); int cap; Node head,tail;
    public int Get(int key) {
        if(!map.ContainsKey(key)) return -1;
        var node = map[key]; Remove(node); AddToHead(node); return node.val;
    }
}

Why SDE-1 gets this: Tests hash map + linked list combo. Amazon loves cache questions.
Follow-up: Thread-safe? Answer: Wrap with lock or use ConcurrentDictionary + careful ordering.

Question: Why does this deadlock in ASP.NET? var result = GetDataAsync().Result;

Expected Answer: ASP.NET has SynchronizationContext. .Result blocks thread waiting for task. Task wants to resume on same context when await completes. But context thread is blocked = deadlock. Console apps don't have SyncContext, so no deadlock.
Fix 1: await GetDataAsync(); all the way up.
Fix 2: .ConfigureAwait(false) in library code to not capture context.
Why they ask: #1 production bug for junior devs. Shows you understand threading model.

Question: Reverse a singly linked list. Iterative + recursive.

Iterative O(n) time O(1) space:

ListNode prev = null, curr = head;
while(curr!= null) {
    var next = curr.next; curr.next = prev;
    prev = curr; curr = next;
}
return prev;

Follow-up: Reverse in groups of K. Amazon L4 asked this in 2026. Need dummy node + careful pointer juggling.
Trap: Losing next pointer before reassigning. Draw boxes.

Question: Concatenate 10k strings. Use string + or StringBuilder? Why?

Answer: StringBuilder. string is immutable. s += "x" creates new string each time = O(n^2) total, lots of GC. StringBuilder.Append uses mutable buffer = O(n).
Numbers: 10k concats: string = 200ms+, StringBuilder = 1ms.
Amazon follow-up: When is string + OK? Answer: Compiler optimizes "a"+"b"+"c" to one concat. But in loops, always StringBuilder.

Question: Implement binary search. Return index or -1.

Key points SDE-1 misses:

int l=0, r=nums.Length-1;
while(l <= r) { // <= not <
    int mid = l + (r-l)/2; // Prevent overflow, not (l+r)/2
    if(nums[mid]==target) return mid;
    else if(nums[mid] < target) l = mid+1;
    else r = mid-1;
}

Traps: 1. mid = (l+r)/2 overflows for large arrays. 2. while(l<r) misses last element. 3. Not handling duplicates - return any index is fine unless asked for first/last.

Question: When to use interface vs abstract class in C#?

Answer:
Interface: Contract only. Multiple inheritance. No state. Use for "can-do" capability: IEnumerable, IDisposable. Since C#8 can have default methods, but still no fields.
Abstract class: Can have state, constructors, partial implementation. Single inheritance. Use for "is-a" with shared code: Stream base class.
Amazon example: ILogger = interface, because Serilog/NLog/Log4Net all implement. DbContext = abstract class, because EF gives you base behavior.

Question: "Tell me time you used data to make a decision"

SDE-1 STAR with metrics:
S: Our college hackathon API for food delivery was timing out at 50 users.
T: I needed to find root cause before 200-user demo.
A: Added Application Insights. Saw 90% time in SELECT * FROM Orders WHERE UserId = @id with no index. Created index: CREATE INDEX IX_Orders_UserId. Also saw EF N+1: fixed with .Include(o => o.Items).
R: p95 latency 2.1s → 95ms. Handled 300 concurrent users in demo. Won 1st place.
Key: Metrics, tool names, specific code change. "We optimized" = reject.

Question: What happens: int i = 5; object o = i; int j = (int)o;

Answer: Boxing: Value type int copied to heap, object reference created. Unboxing: Cast back, copies heap value to stack. Perf cost + GC pressure.
Where it hurts: ArrayList boxes everything. Use List<int> generic to avoid.
Amazon follow-up: How to avoid? Answer: Generics, Span<T>, struct with no boxing.

Question: Given string, find index of first non-repeating char. "leetcode" → 0 for 'l'.

Answer: Two pass with dictionary. Or int[26] for lowercase only.

int[] count = new int[26];
foreach(char c in s) count[c-'a']++;
for(int i=0; i<s.Length; i++)
    if(count[s[i]-'a']==1) return i;
return -1;

Follow-up: Stream of chars? Use Queue + HashMap to track. Amazon asked this 2026.

Question: Difference: AddTransient vs AddScoped vs AddSingleton?

Answer:
1. Transient: New instance every time. Use for lightweight stateless: IEmailService.
2. Scoped: One per HTTP request. Use for DbContext. New request = new DbContext.
3. Singleton: One for app lifetime. Use for IConfig, caches. Careful: can cause memory leaks.
Amazon trap: Injecting Scoped DbContext into Singleton service = captive dependency, crashes. Interviewer tests this.

Question: "Time you took calculated risk"

Answer framework: Amazon loves "I shipped in 2 days vs 2 weeks" stories.
S: Payment service was down, losing $10k/hour. Team wanted 1 week to rewrite.
T: I proposed 4-hour hotfix to unblock, then proper fix later.
A: Added circuit breaker + fallback to cached rates. Deployed behind feature flag. Monitored.
R: Revenue restored in 3 hours. Zero customer impact. Later did full rewrite. Learned: bias for action > perfect code when $ burning.
Don't say: "I waited for manager approval" - Amazon wants ownership.

Question: Your API memory grows 100MB/hour. How to debug?

SDE-1 Answer:
1. dotnet-counters: Check Gen2 heap size, GC count. If Gen2 growing = leak.
2. dotnet-dump: Take heap dump, open in Visual Studio or dotnet-gcdump.
3. Common leaks: Static events not unsubscribed, HttpClient not disposed, CancellationTokenSource not canceled, EF DbContext not disposed.
4. Fix: Use using or IAsyncDisposable. For events: -= in Dispose.
Why they ask: Even L4 should know basic prod debugging. Amazon fires you if you OOM prod.