Privacy
49 bites tagged Privacy — interview questions with model answers, and 60-second explainers.
Privacy compliance in a participant management system
Versioned consent records, data classification and minimization, anonymization or pseudonymization, deletion workflow across stores and backups. Privacy-by-design engineering.
Anonymization vs pseudonymization and key handling
Anonymization is irreversible and exits GDPR scope; pseudonymization is reversible via a separated key; secure that key in a KMS with strict access. precise grasp of two privacy techniques and key management.
What is PII and how to keep it out of logs
Define PII, redact or mask at the logging boundary, and avoid logging sensitive fields at the source. understanding PII plus a concrete log-hygiene strategy.
Epsilon in differential privacy and its trade-offs
Epsilon is the privacy budget bounding how much one record can change outputs; smaller epsilon means stronger privacy but more noise and lower accuracy. privacy-utility-cost balance.
Preventing PII in LLM outputs: curation, fine-tuning, or guardrails
Favor post-processing guardrails as the enforceable last line, backed by data curation; note each layer's tradeoffs and that defense-in-depth is best. Choosing the right layer for PII control.
Preprocessing conversations to protect privacy before fine-tuning
Detect and redact PII with NER plus regex, choose redaction versus pseudonymization, and validate recall. Privacy-preserving data pipelines for training.
Differential privacy vs utility in LLM fine-tuning
Clipping plus calibrated noise per step, smaller epsilon means stronger privacy but degraded accuracy, tracking the privacy budget across epochs. understanding DP-SGD's noise-for-privacy bargain.
Audit an ML pipeline for GDPR compliance
Inventory data and check minimization, verify processing matches stated purpose, build lineage to trace any prediction's inputs. applying GDPR principles technically.
Federated learning architecture, risks, and defenses
Devices train locally and send updates not data, a server aggregates; gradients still leak, enabling inversion or membership inference; defend with secure aggregation and DP. privacy-preserving distributed training.
Enforcing GDPR Right to be Forgotten at scale
Map the subject's data, then crypto-shred the lake, DELETE in the warehouse, and evict caches via an auditable, idempotent workflow. Privacy engineering across stores. Assuming one DELETE suffices everywhere.
Attribute a mobile install to a desktop ad
Deterministic matching via a shared login is accurate but needs auth on both ends; probabilistic fingerprinting scales without login but is noisy and privacy-fraught. cross-device attribution methods and their trade-offs.
UK digital ID plan ends online anonymity for adults
UK plan forces adults to verify social media access with passports or credit cards, ending online anonymity. Engineers must now design age-gating and digital ID systems amid institutional child protection failures and record-low 30% trust in traditional media.
How would you track clicks on a new 'Export Data' button?
Tests prod analytics design beyond console.log. Outline: typed event schema, client validation, beacon or queue with batching, and queryable storage. Red flag: fire-and-forget tracking without retry, sampling, or privacy checks.
How would you secure a moderated screen-sharing session end to end?
Cover encrypted WebRTC transport, KMS-backed storage, access controls, PII redaction, and retention policies. Privacy architecture for sensitive streaming data and stored artifacts.
How can Private Set Intersection enable joint research without exposing raw lists?
Tests cryptography for cross-party data sharing. Strong answers describe PSI's encrypted intersection, cite homomorphic encryption or oblivious transfer, and note key management and compute cost.
How do you reliably deliver a survey to 10,000 users?
Tests systems thinking for outreach at scale. Strong answers cover idempotent delivery, rate-limited async sending, PII handling and retention, plus observability into delivery and response rates.
Design a system integrating analytics data with participant recruiting
Tests system design for behavioral targeting in research ops. A strong answer covers: a warehouse-to-tool pipeline; consent and privacy gates; behavioral SQL segmentation; and frequency capping. Red flag: skipping GDPR and consent to focus only on data joins.
What technical instrumentation supplements a two-week diary study?
Log anonymized feature frequency, session duration, and device switches; link via participant ID. Pairing diary entries with telemetry to triangulate longitudinal UX insights.
Design a cross-product user journey data architecture
This tests portfolio-scale data platform design. A strong answer outlines streaming event ingestion, a unified identity graph, consent-aware PII vaults, and schema registries with data contracts.
Design an ML workflow that masks PII from scientists
This tests privacy-preserving pipeline design and least-privilege access for ML teams. Propose automated de-identification before experimentation, restrict re-identification to production jobs, and enforce role-based access with audit logs.
Apple's Natural Language Framework
Apple's Natural Language Framework turns raw strings into structured meaning on-device without network calls. Tokenize queries or extract entities from user text locally. It is not infallible; heavy synchronous tagging on the main thread freezes your UI.
What data and approach for a simple 30-day DAU forecast?
Tests forecasting from sessionized logs without overengineering. Cite timestamped events, a 30 min session rule, and a regression baseline with day-of-week, recent totals, scored with MAE. Red flag: deep learning before a baseline or ignoring privacy hashing.
Describe the end-to-end data flow for tracking a 'Share' button click
Payload carries event type, user ID, timestamp, device, content; client batches with retry; backend validates and lands in a partitioned store. Full analytics pipeline design. Raw PII, no timestamps, no dedup.
How does Rust differentiate unit and integration tests?
Tests Rust test layout and privacy. Unit tests sit in src/ under #[cfg(test)] and call private functions via super::. Integration tests go in tests/ as external crates. Wrong: claiming it blocks private testing or merging them into src/.
Get Privacy bites daily.
Five a day, five minutes, offline. With quizzes so it sticks.
Open testing — you’ll join as an early tester.