Competition for Amazon’s New Grad (NG) jobs reaches a fever pitch in 2025. According to feedback from posts related to Amazon ng One-third of an Acre, this year’s OA shows a significant trend: it no longer pursues the extreme difficulty of LeetCode Hard, but attaches great importance to the robustness of the code (Robustness) and the processing of boundary conditions (Edge Cases).
Recently, one of our students flawlessly passed two coding questions on the HackerRank platform in 13 minutes. This is not luck, but a precise dimensionality reduction attack on Amazon’s Pattern Recognition. This article will review this set of questions in depth and share the underlying logic for efficient problem solving.
Amazon OA Examination Form and Scoring Criteria
Amazon's OA process usually consists of three parts, of which coding is the first level and the hardest threshold.
- Coding Assessment (70 mins): 2 programming questions.
- Expert Insight: Simply Pass Test Cases is not enough. Amazon's backend scoring system checks your code complexity, variable naming conventions, and whether there is redundant logic.
- Work Simulation (SJT): Multiple choice questions simulating real work scenarios.
- Work Style Assessment: Personality test, the core test points are Leadership Principles (LP).
Amazon OA Overall difficulty & experience
- Platform:HackerRank
- Question volume:2 programming questions
- Total duration:70 minutes
- My time: 13 minutes (T1 8 minutes + T2 5 minutes)
- Difficulty judgment：Easy-Medium, much simpler than the old version of OA that everyone spreads on one-third of an acre of land
- Test center direction：Interval merge, greedy, sliding window, modulo allocation
Amazon's current NG OA style is more "can you write the basic code correctly and handle the boundaries well", rather than the kind of tricky high-frequency questions that get stuck in your mentality.
T1 — Interval Connectivity Reduction
Problem description: Given n intervals [start_i, end_i], you can add up to k new intervals (the length of each new interval does not exceed limit). Find the minimum number of connected blocks (that is, independent interval segments after merging) required.
Core idea:
- First sort all intervals by start and merge overlapping/adjacent intervals (classic Merge Intervals).
- After merging, multiple independent segments are obtained, and the gap between adjacent segments is calculated.
- Sort all gaps, use sliding window or greedy method to cover the largest gaps with up to k intervals of length ≤ limit, until connected blocks can no longer be reduced.
Sample code (Python):
Def minimize_components(intervals, k, limit):
intervals.sort(key=lambda x: x[0])
merged = []
for interval in intervals:
if not merged or merged[-1][1] 0 and i < len(gaps):
if gaps[i] <= limit * k:
k -= (gaps[i] + limit - 1) // limit #How many intervals are needed to cover this gap
i += 1
else:
break
return len(merged)-iCommon error-prone areas: first, forgetting to sort by the left endpoint before merging the range, resulting in a merge error; second, writing start [i+1] and end [i] backwards when calculating gaps; third, the left pointer of the sliding window is not advanced correctly, creating an infinite loop; fourth, when judging the coverage length, the "equal to k" situation is missed.
T2 — Maximizing Distinct Hash Values
Problem description: Given an array of params (the weight of each position), you need to construct a current value current, accumulated by current = (current + i * params[i]) % MOD. Find the maximum number of different hash values that can be obtained.
Core idea:
- The final hash value set size depends on the difference in the modulo result after each accumulation.
- To maximize the number of different values, the accumulation process needs to produce as many new values as possible.
- Greedy strategy: Sort params from large to small, giving priority to large items that can "jump out" of the existing modulus range faster.
Sample code (Python):
Def max_distinct_hashes(params):
params.sort(reverse=True)
seen = set([0])
current = 0
for p in params:
current = (current + p) % (10**9 + 7) # MOD is usually large
seen.add(current)
returnlen(seen)Common error-prone areas: First, the param array is not sorted, and the greedy logic cannot take effect; second, it is not considered that when param [i] is equal to 1, the modulo can only get 0, and other values cannot be generated; third, the growth logic of current_val is written incorrectly; fourth, the "maximum number of different hash values" is misunderstood as the number of elements in the param array after deduplication.
Although this set of questions is not difficult, for students who have never solved questions like Interval or Constructive Greedy, it is still a lot of pressure to finish it within 70 minutes and successfully debug it. If you still have no idea about OA, or you are about to accept the HackerRank link and don’t dare to click on it, see below
Amazon Leadership Principles (LP) – The overlooked “hidden killer”
Many students with excellent skills have failed in Work Simulation after Coding. Amazon's OA is not just a code test, but also a test of whether you are in line with Amazon's values.
Key points for exam preparation:
- Customer Obsession: When answering multiple-choice questions, always put customer interests above KPIs.
- Bias for Action: When information is incomplete, do you choose to wait for perfect data, or act first based on existing data? Amazon prefers the latter (but only if the risks are manageable).
- Ownership: Instead of saying "that's not my job," show responsibility for the entire project.
Final season lifeline! Can’t figure out the CS assignment/Project? try to find Programhelp
Want to overtake in the fierce job hunting season?Programhelp Provide you with full-process hard-core assistance from OA to Offer.
Against OA , we use ToDesk remote invisible operation, support all platforms such as HackerRank and Codesignal, and promise a 100% pass rate, safety and concealment, but there is no charge.
Exist VODuring the interview, we provide flexible solutions: North American CS experts provide real-time idea prompts (covering Code, BQ, and System Design) through self-developed invisible documents to help you answer questions fluently; we also provide proxy interview services based on cameras and voice-changing technology that support "lip-syncing" or "full replacement" modes to perfectly solve the stage fright problem.
There is also a full-service offer package, covering all aspects from OA, interviews to salary negotiations, providing one-stop service until you get a satisfactory offer. Leave the professional work to us, you just need to prepare to join the company!
The job search in NG is really cruel, everyone is grabbing the same opportunities.
But the competition this time is not about who has more difficult questions, but who is more stable, faster and makes fewer mistakes.
So if you’re stressed out and anxious right now – don’t worry, we’ve all been there.
Step by step, stabilizing OA is the best start.
Hope you get the "Congratulations!" email soon.
