Interview report — sample · easyinterview.life

verdict

Explained both problems back precisely before touching the editor, reached the optimal LRU structure on the whiteboard before writing a line, and coded cleanly under live probing. Lost ground on an eviction off-by-one and a partial answer on lock granularity in rapid-fire . Clears the SDE 2 coding bar with room to spare.

rubric breakdown

Problem solving

4/5

Derived the optimal structure from first principles when OrderedDict was challenged.

Coding & correctness

4/5

Both solutions passed dry-run; one boundary bug fixed in 80 seconds once surfaced.

Data structures & complexity

5/5

Instant, correct amortized analysis; explained why the sentinel trick removes branches.

Edge cases & testing

3/5

Both deductions were edges she didn't test unprompted — a consistent, coachable pattern.

Communication

4/5

Narrated while typing; locked the plan verbally before writing code both times.

LRU cache with O(1) operations

mediumreached optimal

presented · solve timer started — stops at passing dry-run

18m 30s

solve time

each segment is clickable — jumps the recording to that phase

Problem explanation3m 30sRestated precisely; surfaced the get-bumps-recency subtlety unprompted✓ clear

Approach — whiteboard5m 30sSketched hash map + doubly linked list; agreed optimal before coding✓ optimal first

Coding · 2 live probes13m 00sClean structure; one eviction off-by-one found under probing, fixed in 80s−1 edge case

final submission · python 3.12 · dry-run passed● probed line● deduction line

1class LRUCache:

2 def __init__(self, capacity: int):

3 self.cap, self.map = capacity, {}

4 self.head, self.tail = Node(), Node() # sentinels

5 self.head.next, self.tail.prev = self.tail, self.head

7 def get(self, key: int) → int:

8 if key not in self.map: return -1

9 node = self.map[key]

10 self._unlink(node); self._push_front(node)

11 return node.val

13 def put(self, key: int, value: int) → None:

14 if key in self.map:probe ① 15:40

15 self._unlink(self.map.pop(key))

16 elif len(self.map) == self.cap:probe ② 20:10

17 self._evict_lru() # ← fixed: evict BEFORE insert

18 node = Node(key, value)

19 self.map[key] = node; self._push_front(node)

probe ① — agent highlighted L14–15

Asked: “Walk me through the overwrite path — same key, new value.”

Answered immediately and correctly: pop-and-relink keeps size stable, no eviction fires. Passed unaided.

probe ② — deduction −1 · L16–17

Asked: “What happens at exactly capacity — insert then evict, or evict then insert?”

Original code evicted after insert, briefly holding cap+1. Found and fixed it in 80s once probed — but didn't catch it on her own dry-run. −1 edge cases.

Rapid-fire · 3 × 1 min3m 30s2 clean, 1 partial — lock granularity didn't go past a global lock2/3 + partial

Code trace3m 30sTraced put("d",4) at capacity — 7 of 8 steps right, one pointer slip self-corrected7/8 steps

Insert interval into sorted, disjoint list

mediumO(n) single pass

presented · skipped whiteboard — went straight to code after a verbal plan

9m 00s

solve time

Coding · 1 probe9m 00sThree-phase merge, right first try; empty-list edge only after probe−1 edge case

proctoring & anomalies

Continuous camera + screen + audio monitoring across 46:00. Three low-severity signals were raised and auto-reviewed — none indicate assistance or a policy breach.

0 need review

camera

Gaze left frame for 16s

auto-cleared

Eyes tracked down-left during a pause in narration, coinciding with reading the prompt. Returned to screen; no device or second face detected in the window.

screen

Clipboard paste into editor blocked

informational

A single 42-character paste was blocked by the editor's paste guard; the candidate retyped the line manually. Logged for transparency, not penalised.

audio

Brief second voice at 22:40

auto-cleared

~3s of background speech, not responsive to the interviewer's question and outside the candidate's lip movement — consistent with household noise.

strengths

+ Locks the approach with the interviewer before coding — zero rewrites

+ Explains code as she writes it; probes rarely surprise her

+ Traces her own code accurately under pressure

growth areas

− Boundary cases surface only when probed — both deductions were edges

− Concurrency answers stop at the first workable idea (global lock)

communication signals

2.1

fillers / minute

clarifying questions asked

80s

avg. recovery after a probe

7.5

composure — “calm, revises openly”

Narrated continuously through both solves; silence only during the trace, which is expected.

Priya Sharma

LRU cache with O(1) operations

Insert interval into sorted, disjoint list