Technische Interviews als Senior Engineer meistern

// A framework for approaching system design interviews
interface DesignFramework {
  step: string;
  timeAllocation: string;
  activities: string[];
}
 
const systemDesignFlow: DesignFramework[] = [
  {
    step: "Requirements Clarification",
    timeAllocation: "5-7 minutes",
    activities: [
      "Identify functional requirements (what the system does)",
      "Identify non-functional requirements (scale, latency, consistency)",
      "Clarify scope: what's in, what's out",
      "Establish rough numbers: DAU, QPS, storage",
    ],
  },
  {
    step: "High-Level Design",
    timeAllocation: "10-12 minutes",
    activities: [
      "Draw core components and their interactions",
      "Identify data flow: write path and read path",
      "Choose communication patterns: sync, async, event-driven",
      "Call out key technology choices and WHY",
    ],
  },
  {
    step: "Detailed Design",
    timeAllocation: "12-15 minutes",
    activities: [
      "Deep dive into 2-3 critical components",
      "Data model and schema design",
      "API design for key endpoints",
      "Caching strategy and invalidation",
    ],
  },
  {
    step: "Scaling and Tradeoffs",
    timeAllocation: "8-10 minutes",
    activities: [
      "Identify bottlenecks under load",
      "Horizontal scaling strategy",
      "Database partitioning / sharding approach",
      "Failure modes and mitigation",
    ],
  },
];

// Common estimation building blocks
const estimationCheatSheet = {
  storage: {
    textMessage: "~100 bytes",
    tweet: "~300 bytes (with metadata)",
    photo: "~200 KB (compressed)",
    video1Min: "~10 MB (compressed)",
    userProfile: "~1 KB",
  },
  throughput: {
    ssdRead: "~100K-500K IOPS",
    hddRead: "~100-200 IOPS",
    networkWithinDC: "~1-10 Gbps",
    databaseQuerySimple: "~10K QPS per node",
    cacheRead: "~100K-1M QPS (Redis)",
  },
  latency: {
    l1Cache: "~1 ns",
    ram: "~100 ns",
    ssdRandom: "~100 μs",
    networkWithinDC: "~0.5 ms",
    networkCrossContinent: "~100 ms",
    databaseQuery: "~1-10 ms",
  },
  scale: {
    secondsPerDay: 86400,
    secondsPerMonth: "~2.5M",
    bytesPerGB: "~1e9",
    bytesPerTB: "~1e12",
  },
};
 
// Example estimation: "Design a system that handles 100M DAU"
function estimateLoad(): void {
  const dau = 100_000_000;
  const actionsPerUser = 10;
  const dailyActions = dau * actionsPerUser; // 1 billion/day
  const qps = dailyActions / 86400; // ~11,574 QPS
  const peakQps = qps * 3; // ~35K QPS (3x peak multiplier)
  const storagePerAction = 500; // bytes
  const dailyStorage = dailyActions * storagePerAction; // 500 GB/day
  const yearlyStorage = dailyStorage * 365; // ~180 TB/year
}

// ❌ Bad: Vague, unfocused answer
const badAnswer = {
  question: "Tell me about a time you disagreed with a technical decision",
  response:
    "I disagreed with my manager about using microservices. " +
    "I thought we should use a monolith. We discussed it and " +
    "eventually went with microservices. It worked out okay.",
};

// ✅ Good: Structured STAR-L response
interface BehavioralResponse {
  situation: string;
  task: string;
  action: string;
  result: string;
  learnings: string;
}
 
const strongAnswer: BehavioralResponse = {
  situation:
    "Our team of 4 was building a new payment processing service. " +
    "The tech lead proposed splitting it into 6 microservices from " +
    "day one. We had no existing infrastructure for service orchestration.",
  task:
    "I needed to advocate for a simpler architecture without " +
    "undermining the tech lead's authority or creating team friction.",
  action:
    "I wrote a one-page RFC comparing both approaches with specific " +
    "tradeoffs: deployment complexity, latency overhead, debugging " +
    "difficulty. I proposed starting as a modular monolith with clean " +
    "module boundaries that could be extracted later. I presented it " +
    "as 'how do we get to market fastest with the option to split later' " +
    "rather than 'your idea is wrong.'",
  result:
    "We shipped the monolith in 3 months instead of the estimated 6 " +
    "for microservices. After 8 months, we extracted the notification " +
    "module into its own service when load patterns justified it. " +
    "The other modules stayed monolithic.",
  learnings:
    "Framing matters more than being right. Writing the RFC forced me " +
    "to quantify my intuition, which made the conversation productive " +
    "instead of opinion-based. I also learned that 'defer the decision' " +
    "is often the best architecture advice.",
};

// The approach matters more than speed
// Demonstrate: problem decomposition, edge case handling, testing mindset
 
// Example: Design a rate limiter
interface RateLimiterConfig {
  maxRequests: number;
  windowMs: number;
}
 
class SlidingWindowRateLimiter {
  private windows: Map<string, number[]> = new Map();
  private config: RateLimiterConfig;
 
  constructor(config: RateLimiterConfig) {
    this.config = config;
  }
 
  isAllowed(clientId: string): boolean {
    const now = Date.now();
    const windowStart = now - this.config.windowMs;
 
    // Get or initialize request timestamps for this client
    let timestamps = this.windows.get(clientId) || [];
 
    // Remove expired timestamps
    timestamps = timestamps.filter((t) => t > windowStart);
 
    if (timestamps.length >= this.config.maxRequests) {
      this.windows.set(clientId, timestamps);
      return false;
    }
 
    timestamps.push(now);
    this.windows.set(clientId, timestamps);
    return true;
  }
 
  // Production consideration: memory cleanup
  cleanup(): void {
    const now = Date.now();
    for (const [clientId, timestamps] of this.windows) {
      const valid = timestamps.filter(
        (t) => t > now - this.config.windowMs
      );
      if (valid.length === 0) {
        this.windows.delete(clientId);
      } else {
        this.windows.set(clientId, valid);
      }
    }
  }
}
 
// Verbalize tradeoffs while coding:
// "I'm using a sliding window for accuracy. A fixed window would be simpler
// but allows 2x the rate at window boundaries. The tradeoff is memory —
// we store individual timestamps instead of a counter."

const seniorQuestions: string[] = [
  // Engineering culture
  "How do architectural decisions get made here? Is there an RFC process?",
  "What does the on-call rotation look like, and how are incidents handled?",
 
  // Team dynamics
  "How much autonomy do engineers have in choosing technical approaches?",
  "What's the ratio of planned work to interrupt-driven work?",
 
  // Growth and impact
  "What does a successful first 90 days look like for this role?",
  "Can you describe a recent technical decision the team made that was controversial?",
 
  // Organizational health
  "How does the engineering team handle technical debt?",
  "What's the deployment frequency, and what does the release process look like?",
];