Capital One Senior Software Engineer VO Real Questions: System Design, Power Day, and LP

Capital One's Senior SWE bar is significantly higher than entry-level or TDP (Tech Development Program) - especially Power Day, where 4 back-to-back rounds compress algorithm + system design + Java/Spring engineering practice + Behavioral into a single grueling day. Community descriptions of Power Day are often vague. This article gives the real per-round questions and answer skeletons based on a Senior SWE pass case.

This article does not cover entry-level OA (the site already has a SWE TDP OA piece). It focuses on Senior-tier system design complexity, Java memory model follow-ups, and surgical hits on Capital One's five Leadership Principles.

Senior SWE Pipeline

Stage	Length	Format	Decision weight
Recruiter Screen	30 min	Why Capital One + resume	Entry
Tech Phone	60 min	LeetCode Medium + light system	Critical filter
Power Day Round 1	60 min	Coding Hard	Decisive
Power Day Round 2	60 min	System Design	Decisive
Power Day Round 3	60 min	Java + Spring engineering	Decisive
Power Day Round 4	45 min	Behavioral / LP	Decisive

Power Day rule: at least 3 of 4 rounds at strong hire or hire to extend an offer. Any single strong-no is essentially a veto.

Recruiter Screen: Five Questions in 30 Minutes

1. "Walk me through your last 5 years of work."
2. "Why are you interested in Capital One specifically?"
3. "What's your salary expectation? Can you work in McLean / Plano / NYC?"
4. "Describe a project where you owned end-to-end delivery."
5. "Do you have any current offers?"

Answer strategy:

• "Why Capital One" must mention fintech transformation: Cloud-First, full AWS migration, ML in Underwriting.
• Salary expectations: a band (e.g., 240-290k base + 20% bonus). Never a single number.
• Offers: be honest, but do not name specific companies.

Tech Phone: Standard LeetCode Style

Real question: Top K Frequent Words with Tie-Break

Given a log array, return the K most frequent IP addresses. On ties, sort lexicographically.

import heapq
from collections import Counter

def top_k_ips(logs, k):
    cnt = Counter(logs)
    heap = [(-c, ip) for ip, c in cnt.items()]
    heapq.heapify(heap)
    return [heapq.heappop(heap)[1] for _ in range(k)]

Complexity: O(N log N), N is the count of unique IPs. Follow-up: K=10 with N=1e6 - optimize? Answer: a min-heap of size K, O(N log K).

Power Day Round 1: Coding Hard

Real question: Distributed Rate Limiter

Build a distributed rate limiter: per user_id, at most 100 requests in any rolling 60-second window. Multiple instances must share state.

Approach: rolling window backed by a Redis sorted set.

import time
import redis

def rate_limit_check(r: redis.Redis, user_id: str, limit=100, window=60):
    key = f"rl:{user_id}"
    now = time.time()
    pipe = r.pipeline()
    pipe.zremrangebyscore(key, 0, now - window)
    pipe.zcard(key)
    pipe.zadd(key, {str(now): now})
    pipe.expire(key, window)
    _, count, _, _ = pipe.execute()
    return count < limit

Follow-up chain:

1. Redis SPOF? -> Redis Cluster + consistent hashing.
2. Network partition - how do we degrade? -> local token bucket fallback.
3. Redis becomes the bottleneck under spikes? -> Lua script for atomic ops + sharding by user_id.

Power Day Round 2: System Design

Real question: design Capital One's real-time fraud detection engine

Requirements:

• 50K credit-card transactions per second
• end-to-end latency < 100 ms
• false positive rate < 0.1 percent
• decisions powered by historical rules plus an ML model

Answer skeleton (5-layer architecture):

1. Ingestion: Kafka, partitioned by user_id for in-order per-user delivery.
2. Feature Store: Redis (real-time + 60s rolling) plus Cassandra (30-day history).
3. Decision Engine:
   • Rules engine (Drools or in-house DSL)
   • ML model (XGBoost online inference via TensorFlow Serving)
   • Both paths run in parallel -> voting or weighted fusion
4. Storage: transaction outcomes plus decision traces in DynamoDB (write-heavy).
5. Feedback Loop: agent-flagged false positives -> retraining dataset.

Critical tradeoffs:

• 100ms hard SLA: ML inference must finish under 30 ms, otherwise serve a cached score.
• Consistency vs availability: finance is CP over AP.

Power Day Round 3: Java + Spring

Real question: Spring Boot microservice debug

Given controller -> service -> repository code, production sees occasional ConcurrentModificationException. Locate and fix.

Signals being tested:

• ConcurrentHashMap vs Collections.synchronizedMap differences
• Whether Spring @Transactional propagation swallows exceptions
• ThreadLocal context loss in async / WebFlux

Typical fix

// Bad: HashMap dies under multi-thread access
private Map<String, Account> cache = new HashMap<>();

// Good: thread-safe
private Map<String, Account> cache = new ConcurrentHashMap<>();

// Better: size limit + LRU
private Map<String, Account> cache = Collections.synchronizedMap(
    new LinkedHashMap<>(1000, 0.75f, true) {
        @Override
        protected boolean removeEldestEntry(Map.Entry<String, Account> eldest) {
            return size() > 1000;
        }
    }
);

Follow-up: when do you reach for ConcurrentHashMap.compute() over putIfAbsent? Answer: when you need atomic read-modify-write.

Power Day Round 4: Behavioral / LP

Capital One's Five LPs

1. Customer Obsession
2. Champion Diversity, Inclusion & Belonging
3. Do the Right Thing
4. Lead with Excellence
5. Take Smart Risks

Prepare 1-2 STAR stories per LP. Senior SWE STAR must include quantified results:

• "Reduced API latency by 40% (P99 from 250ms to 150ms)"
• "Migrated 12 microservices to AWS, saving $200K annually"
• "Mentored 3 junior engineers, all promoted within 18 months"

Sample real question

"Tell me about a time you took a smart risk that didn't work out."

STAR skeleton:

• Situation: project context plus the risk
• Task: your role and decision authority
• Action: what you did and why
• Result: it failed, but what you learned and how you applied that lesson afterward

Key: Capital One does not punish failure - they evaluate the depth of reflection.

FAQ

Q1: Are the 4 Power Day rounds in one day or spread out? Usually all in one day: 4 x 60 minutes plus 1 x 45 minutes, around 5 hours total, with a 30-minute lunch break.

Q2: Is Java required for Senior SWE? Almost always. Capital One backend leans Java + Spring Boot, with some Cloud teams on Python. Power Day Round 3 typically mandates Java.

Q3: Must System Design use AWS? Not required, but strongly recommended. Capital One is AWS-first; answers using GCP or Azure invite repeated follow-ups.

Q4: Salary base range? Senior SWE base is usually 200-280k. With stock + bonus, total comp is 280-380k (McLean / NYC slightly above Plano).

Q5: Visa friendly? Senior roles are H1B-friendly, but some Cloud teams restrict to US citizens or GC due to contract terms. Confirm with the recruiter directly.

---

Preparing for Capital One Senior SWE Power Day?

If you want a 5-layer System Design walkthrough, 4-round Power Day mock, or a real person doing VO proxy / VO assist live shadowing on Power Day with Java concurrency support, we can talk through a complete OA proxy / VO assist / VO proxy plan.

---

Contact

Need real interview questions and a custom prep plan? Add WeChat Coding0201 now to get questions.

Email: [email protected] Telegram: @OAVOProxy