HackWithInfy Round 2 Live Coding round·Engineering·Hard·20 min

Infosys Specialist Programmer Interview — HackWithInfy Round 2 Live Coding

20 min · 1 credit · scorecard at the end
Z
ZeroPitch Research Desk
Updated 2026-05-25 · Newly added · be among the first to take it
Field
Engineering
Company
Infosys
Role
Specialist Programmer (Power Programmer premium track)
Duration
20 min
Difficulty
Hard
Completions
New
Updated
2026-05-25

What this round is about

  • Topic focus. The HackWithInfy Round 2 Specialist Programmer technical loop, covering one hard live-coded DSA problem at optimal complexity using segment tree with lazy propagation, one system design at 1M requests per second with explicit capacity numbers, depth on CS fundamentals across JVM garbage collection and locks vs lock-free and cache eviction, and a culture-fit closing on research aptitude and open-source contribution.
  • Conversation dynamic. Arjun Kumar, a 9-year Principal Programmer in the Infosys Strategic Technology Group at Bangalore Electronic City and a 5-year HackWithInfy SP panel member, is warm at the start and sharp in the middle. He interrupts during live coding on purpose to test whether you can keep your train of thought under verbal pressure. He never praises an answer.
  • What gets tested. Optimal-complexity reasoning under live interruption, capacity-aware system design with named tradeoffs, CS fundamentals depth where you pick one algorithm over another with a real scenario, conviction defence when the panel pushes back on a correct answer, and a real open-source contribution or research paper named in the culture-fit closing.
  • Round format. Twenty minutes, four blocks, candidate-coded on the canvas. Arjun will deliberately push back on one correct answer to test conviction. The whiteboard is on throughout the DSA and system-design blocks.
  • Why this loop is hard. Twenty minutes is tight: roughly seven minutes of live coding on a segment tree, six on a 1M QPS design, four on CS fundamentals depth, and three on culture fit. Arjun controls the pacing and will cut you off the moment the depth flattens. Every block has a content gate, not a time gate.

What strong answers look like

  • Constraint-anchored complexity target. State the input size out loud before picking the data structure. Example phrase: N is up to 1e5, so I am targeting O of N log N. Segment tree with lazy propagation, not Fenwick tree, because the range-update is more natural there.
  • Capacity numbers before architecture. State QPS, latency budget, hot key ratio, and shard key choice before drawing any box. Example phrase: 1M requests per second average, sub 10ms p99 budget for the limiter itself, top 1 percent of users at 10x the average, shard by user ID hash with consistent hashing.
  • Real-scenario CS fundamentals. Pick LRU for a recency-dominated workload like a CDN edge cache and LFU for a frequency-dominated workload like a database buffer pool. Pick G1 for sub-200ms pause budget and ZGC for sub-10ms pause budget. Name the tradeoff with one sentence of reasoning.
  • Named open-source contribution. Name one upstream contribution by repository, file, and the problem fixed. Example phrase: I sent a patch to Apache Kafka, file producer config, fixed a default value that caused message loss under broker failure. Personal portfolio repositories do not count as open-source.
  • Conviction under pushback. When Arjun deliberately challenges a correct decision, defend with one line of reasoning before reconsidering. Example phrase: I picked token bucket because the design allows short bursts up to the bucket capacity. If you want me to optimise for strict per-second steady-state, I would switch to a fixed window counter.

What weak answers look like (and how to avoid them)

  • Brute force on a constrained input. Writing O of N squared on N up to 1e5 reads as a flight risk. Avoid by stating the input size and complexity target out loud before writing a single line of code.
  • System design boxes before capacity numbers. Sketching Redis and gateway boxes without stating QPS or latency budget. Avoid by writing QPS, latency, hot key ratio, and shard key as four bullets on the canvas before any box.
  • Recited CS definitions without scenarios. Reciting LRU is least recently used and LFU is least frequently used without picking one for a real workload. Avoid by preparing one real scenario per algorithm in advance.
  • Capitulating on the first pushback. Immediately switching your answer when Arjun challenges a correct decision. Defend with one sentence of reasoning first. Capitulating reads as low conviction and is the most cited soft-reject signal.
  • Saying you read papers without naming one. The culture-fit closing tests for one specific paper or one upstream commit. Saying I read a lot of papers and blogs without naming a title and a one-line takeaway is the most cited soft-reject signal across HackWithInfy SP debriefs. Have one title, one venue, and one idea on your tongue before you sit down.

Pre-interview checklist (2 minutes before you start)

  • Recall your one open-source contribution. Have the repository name, the file or commit hash, and the one-sentence description of the problem you fixed on your tongue. If you do not have an upstream contribution, have one technical blog URL ready.
  • Have segment tree with lazy propagation in muscle memory. Practice the range-update range-query template once the morning of the interview. Confirm you can write the build, update, and query functions without a reference.
  • Pull up the capacity-number template. Four bullets to write on the canvas before any system-design box: QPS target, latency budget added by the system, hot key ratio for the top 1 percent of keys, shard key choice with reason.
  • Identify one paper you have read in the last 6 months. Have the title, the conference or journal, and the one-sentence main idea ready. Examples: Dynamo by DeCandia et al at SOSP 2007, Spanner by Corbett et al at OSDI 2012, or The Log by Jay Kreps on LinkedIn engineering.
  • Re-read your LRU vs LFU vs ARC vs Clock scenarios. One sentence each on when one beats the others. CDN edge for LRU, database buffer pool for LFU, mixed workloads for ARC, Linux kernel for Clock.
  • Confirm with yourself: yes to 1-year bond at 1L INR, yes to relocation to Bangalore, Pune, Hyderabad, or Mysore. No city preference, no hesitation on the bond.

How the AI behaves

  • Probes complexity on every line of code. Arjun interrupts during live coding to ask for the complexity of the loop you just wrote, on purpose. Have the answer ready as you write.
  • No mid-interview praise. He will not say great answer or exactly. He acknowledges with accha, theek hai, right, go on, and pushes deeper.
  • Deliberate pushback once. He will challenge one correct decision during the round, often on the algorithm choice or the rate-limiter design, to test conviction. Defend with reasoning. Capitulating immediately reads as low conviction.
  • Cross-questions buzzwords. If you say microservices, he asks which two services and why. If you say Redis, he asks which Redis primitive and the failure mode. Speak in specifics, not categories.
  • Names a specific Hindi acknowledgement. Arjun uses accha, theek hai, right, ek minute as fillers between turns. None of these are praise. Read them as continue, not as you are doing well. If he says go on, finish the sentence you started and add the missing primitive or capacity number.

Common traps in this type of round

  • Brute force on a constrained input. O of N squared on N up to 1e5 is the single most cited reject signal on the DSA block. Name the complexity target before writing.
  • Architecture boxes before capacity numbers. Drawing Redis and the gateway without saying 1M QPS or sub 10ms p99 reads as bootcamp-level system design. Write the four capacity bullets first.
  • Sliding window log without naming the memory cost. Picking sliding window log over token bucket without acknowledging that it stores one entry per request per key is the most cited rate-limiter mistake. Name the tradeoff explicitly.
  • Reciting CS definitions without scenarios. LRU is least recently used as the only sentence, no scenario. Prepare one real workload per algorithm.
  • Capitulating on the first pushback. Switching your answer immediately when challenged reads as low conviction. Defend with one sentence first.
  • Generic culture-fit answer. I read research papers without naming one is the most cited culture-fit reject signal. Name one paper by title and conference.
  • Visible hesitation on the bond or relocation. The 1-year bond at 1L INR and relocation to Bangalore or Pune or Hyderabad or Mysore is mandatory. A confident yes is expected.

Sample problems you'll face

The 3 problems below are the same ones you'll work through in the live session — no surprises. Read the constraints carefully; the AI persona will refer you to the on-canvas card by problem number.

  1. 1Maximum Sum Subarray of Size K

    Given an integer array and a positive integer k, return the maximum sum of any contiguous subarray of exactly length k.

    Example inputarr = [2, 1, 5, 1, 3, 2], k = 3
    Example output9 (subarray [5, 1, 3])
    • 1 ≤ k ≤ length(arr) ≤ 10^5
    • Sliding window O(n); naive O(n × k) won't clear time limit
    • Defend against k = length(arr) and k = 1 edge cases before writing
    • Use the on-canvas card to read the prompt; sketch or write code on the whiteboard. The AI sees what you draw.
  2. 2Longest Palindromic Substring

    Return the longest palindromic contiguous substring. If multiple have the same length, return any.

    Example inputs = "babad"
    Example output"bab" (or "aba")
    • 1 ≤ length(s) ≤ 1000
    • Expand-around-center O(n²) acceptable; Manacher O(n) bonus
    • Handle odd-length AND even-length palindromes
    • Use the on-canvas card to read the prompt; sketch or write code on the whiteboard. The AI sees what you draw.
  3. 3Range Sum Query with Point Updates

    Implement a data structure that supports update(i, val) and rangeSum(l, r) operations. Both must run in O(log n).

    Example inputarr = [1,3,5]; update(1, 2); rangeSum(0, 2)
    Example output8 (after update arr becomes [1,2,5])
    • 1 ≤ length(arr) ≤ 10^5, up to 10^5 operations
    • Segment tree OR Fenwick (BIT) — defend your choice
    • Sketch the tree shape for n=8 before writing code
    • Use the on-canvas card to read the prompt; sketch or write code on the whiteboard. The AI sees what you draw.

Interview framework

You will be scored on these 7 dimensions. The full rubric with definitions is below.

Optimal Complexity Reasoning
How rigorously you state the input constraint and target complexity before writing code, and whether your data structure choice matches the input-size hint with a rejected alternative named.
22%
Capacity Aware System Design
How precisely you pin QPS, latency budget, hot key ratio, and shard key as numbered bullets on the canvas before drawing any architecture box for the rate-limited API gateway.
20%
Algorithm Tradeoff Articulation
How clearly you pick one algorithm and name the rejected alternatives with a one-sentence reject reason anchored in memory cost, burst tolerance, or boundary behaviour.
15%
Cs Fundamentals Scenario Depth
How concretely you pick one cache eviction policy or GC algorithm or lock-free primitive for a real workload, not the textbook definition.
13%
Hot Key Handling
Whether you name the top 1 percent user problem unprompted, and whether you propose a specific mitigation with the precision-vs-throughput tradeoff named.
10%
Verbal Articulation Under Interruption
How cleanly you finish your sentence when Arjun interrupts mid-coding, then resume the same function without restarting from scratch or losing the variable you were tracking.
10%
Research And Open Source Signal
Whether you name one paper or one upstream commit by repository and file with the main idea or the problem fixed, in the culture-fit closing.
10%

What we evaluate

Your final scorecard breaks down across these dimensions. The full rubric and tier criteria are revealed inside the interview itself.

  • Optimal Complexity Reasoning22%
  • Capacity Aware System Design20%
  • Algorithm Tradeoff Articulation15%
  • CS Fundamentals Scenario Depth13%
  • Hot Key Handling Depth10%
  • Verbal Articulation Under Interruption10%
  • Research And Open Source Signal10%

Common questions

What does the Infosys Specialist Programmer HackWithInfy Round 2 actually test?
It tests whether you can live-code a hard DSA problem at optimal complexity, design a rate-limited API at 1M requests per second with capacity numbers stated up front, and explain CS fundamentals on locks vs lock-free, JVM garbage collection algorithms, and cache eviction policies with a real scenario. The panel probes verbal articulation under interruption, defends one position when challenged, and asks for a real open-source contribution or research paper read recently. The bar for the 9.5 LPA Specialist Programmer offer is roughly the top 100 nationally out of thousands of HackWithInfy Round 1 finishers.
How should I structure my live coding answer in HackWithInfy Round 2?
State the input constraint out loud first. If N is up to 1e5, target O of N log N, not O of N squared. Name the data structure with a reason for rejecting the two closest alternatives before you write code. Derive the complexity verbally while writing the code, not after. When the interviewer interrupts, finish your sentence then answer their question, then resume from where you stopped. Dry-run on a small input at the end. Brute force solutions on a problem with a clear input-size hint are read as a flight risk.
What are common failure modes in HackWithInfy Round 2 for Specialist Programmer?
Recurring rejection patterns are jumping to code without clarifying constraints, writing brute force when the input size hints at an optimal algorithm, going silent under interruption and losing the train of thought, sketching system design without any capacity numbers or shard key reasoning, reciting LRU vs LFU vs Clock without a real scenario, mixing up isolation level anomalies under follow-up, giving a generic one-line answer on research aptitude, and showing visible discomfort about the 1-year bond or relocation to Mysore.
How do I handle the system design portion of the Specialist Programmer interview?
State capacity numbers before any architecture box. For a rate-limited API at 1M requests per second with per-user quota, name the QPS target, the average and p99 latency budget added by the limiter, the hot key ratio for the top 1 percent of users, and the shard key choice with reason. Pick one algorithm such as token bucket and name two rejected alternatives such as leaky bucket and sliding window log with a one-sentence reject reason per alternative. Then descend into Redis primitives, fail-open vs fail-closed default, and the operational rollout in shadow mode.
What should I prepare for the CS fundamentals portion?
Master four areas at depth. JVM garbage collection algorithms mark-sweep, mark-compact, generational, G1, and ZGC with when to pick which. Locks vs lock-free programming with CAS and the ABA problem, plus a lock-free queue walkthrough. Cache eviction policies LRU, LFU, ARC, and Clock with one real scenario each where one beats the others. Database isolation levels read uncommitted, read committed, repeatable read, and serializable with the specific anomaly each one prevents, dirty read, non-repeatable read, and phantom read.
How do I answer the culture-fit and research aptitude question?
Name one paper you have read in the last six months by title, conference, and one-sentence summary of its main idea. Name one open-source contribution by repository, file, and the problem you fixed, not just I forked the repo. Defend a technical opinion when the panel pushes back with one sentence of reasoning, do not immediately switch your answer. Confirm clearly that you are comfortable with the 1-year bond at 1L INR and with relocation to Bangalore, Pune, Hyderabad, or Mysore. Visible hesitation here is read as a flight risk.
How is the AI interviewer different from a real HackWithInfy Round 2 panel?
The AI mimics a Principal Programmer from the Infosys Strategic Technology Group on the HackWithInfy SP panel. It will not praise your answers or confirm if you are right. It will listen for the specific reasoning beats including constraint statement, algorithm choice with rejection of alternatives, capacity numbers up front, real-scenario CS fundamentals, and a named open-source contribution. It will deliberately push back on one correct answer to test your conviction. Switching answers immediately reads as low conviction.
How is scoring done for this mock interview?
Scoring evaluates six dimensions. Optimal-complexity reasoning under live coding pressure. Capacity-aware system design with named tradeoffs. CS fundamentals depth on locks, JVM, and cache eviction. Verbal articulation under interruption. Conviction defence under panel pushback. Research and open-source signal in the culture-fit closing. You are graded on the specificity of your evidence and the quantitative grounding of your numbers, not on diagram aesthetics or buzzword count. A transcript-backed scorecard names the exact moment your reasoning lost the panel.
What does a strong Specialist Programmer answer sound like?
On DSA, you state N up to 1e5, target O of N log N, name segment tree with lazy propagation, reject Fenwick tree because the range-update is harder there, then derive the complexity verbally while writing code. On system design, you state 1M QPS, sub 10ms p99 budget, top 1 percent of users at 10x the average, shard by user ID hash, then propose token bucket on Redis with Lua scripting for atomicity. On culture fit, you name one paper such as the original Dynamo paper by DeCandia et al at SOSP 2007.
What should I do in the first 2 minutes of the interview?
Greet the panel briefly. When asked about an open-source contribution, name one specific upstream contribution by repository and commit or by file. If you do not have an upstream contribution, name one technical blog post you have written with the URL and the topic. Do not list a personal portfolio repository as an open-source contribution. Then confirm comfort with the 20-minute time budget and the structure of one DSA problem, one system design, some CS fundamentals, and a culture-fit closing. Stay calm, the panel sets the pace.
Is HackWithInfy Round 2 the same difficulty for Tier 2 colleges like SMVEC?
The bar is the same across all colleges, IIT, NIT, BITS, or Tier 2 institutions like SMVEC. The Specialist Programmer offer is roughly 150 to 200 nationally per year out of thousands of HackWithInfy Round 1 finishers, and the panel does not adjust for pedigree. SMVEC candidates who have cleared SP in recent years cited the same preparation roadmap: segment trees and lazy propagation, advanced DP, graph shortest paths, system design with capacity numbers, JVM internals, and at least one named open-source contribution or published technical blog post.

Sources this interview is built on

Real candidate-report URLs (Glassdoor / AmbitionBox / PrepInsta / GeeksforGeeks / Medium) reviewed when authoring the questions, persona, and rubric. Verify the realism yourself.