Problem Statement and Data Classification Mission
Problem Statement and Data Classification Mission — data classification system design interview section.
Problem Statement and Data Classification Mission
Design a data classification platform like Microsoft Purview, Google Cloud DLP, or BigID: continuously discover sensitive data across SaaS, cloud object stores, warehouses, and on-prem shares, assign standardized labels, and push enforcement to DLP, encryption, and access policies.
- Discovery answers where sensitive data lives; classification answers what it is; enforcement answers how it is protected
- Multi-tenant SaaS must never leak scan results or samples across customers
- Interviewers probe catalog graph, classifier accuracy, and policy propagation—not a one-time spreadsheet
Mechanism
Connectors crawl metadata and content samples → normalization to a canonical schema → classifier ensemble (regex, dictionary, ML) → label assignment with confidence → policy engine emits actions to downstream security tools.
Failure drills
If discovery stalls, enforcement drifts stale—treat scan SLO as security SLO. If ML false positives spike, analysts overwhelm and teams disable policies.
Cost and capacity
| Signal | Target |
|---|---|
| Connectors | 120+ types |
| Assets indexed | 50M+ |
| Scan throughput | 8 PB/day |
| Label latency | <15 min P95 |
1 public record AssetRef(String tenantId, String uri, String connector) {}
1 def canonical_uri(raw: str) -> str: 2 return raw.rstrip('/').lower()
1 export interface ClassificationMission { discover: boolean; label: boolean; enforce: boolean; }
Why interviewers care
Data Classification interviews reward crisp scope, explicit trade-offs, and failure stories—not generic microservice diagrams.
Interview checkpoint
Name one failure story for Problem Statement and Data Classification Mission that proves you understand real outages, not happy-path diagrams.
Key Highlights
- •Discovery answers *where* sensitive data lives; classification answers *what* it
- •Metrics: Connectors, Assets indexed, Scan throughput, Label latency
- •Design a **data classification platform** like Microsoft Purview, Google Cloud DLP, or Big
Section Rescue Kit
Buzzwords to use:
Safe statements:
- "For Problem Statement and Data Classification Mission, I separate discovery metadata from enforcement policies so scans can be async."
- "I would quantify false-positive budget before choosing ML-only classifiers."
- "Tenant isolation is enforced on catalog partitions and scan credentials, not UI filters alone."