← Nuro Interview InsightsSoftware Engineer

Nuro·Software Engineer·Onsite - System Design / Architecture·Staff

StaffPrefer not to say
Jun 2026

Summary

Nuro system design round, one big question that basically ate the whole session. The scope was massive and I don't think I covered everything they wanted, but it was a genuinely interesting problem.

Questions Asked (1)

Q1

Design a system that ingests large volumes of raw autonomous vehicle logs (sensor data, perception outputs, video) and allows engineers to efficiently search for specific scenarios, such as cases where the AV encountered a pedestrian or a particular traffic sign. Cover ingestion, scene tagging, indexing, query API, petabyte-scale storage tiering, and access control.

System DesignTechnical Trade-offsData Modeling
Author's notes

This one is deceptively wide.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements (data volume, query patterns, latency, retention) and then design a pipeline that decouples ingestion from processing using a message queue. Propose a multi-tier storage architecture with hot/warm/cold layers and an indexing strategy that supports efficient scenario search. Finally, address access control and trade-offs between cost, latency, and complexity.

Pro tip: Emphasize the importance of metadata and automated tagging early in the pipeline to enable efficient search without reprocessing raw data. Also, discuss how you would handle data skew and ensure scalability as data grows to petabytes.

1. Clarify Requirements and Scope

Ask questions to understand data volume (e.g., petabytes per day), query types (e.g., pedestrian detection, traffic sign recognition), latency requirements (real-time vs batch), retention policies, and access control needs.

2. Design Ingestion Pipeline

Propose a scalable ingestion layer using a distributed message queue (e.g., Kafka) to handle high throughput. Ingest raw logs and sensor data, then process them asynchronously with stream processing (e.g., Flink) to extract metadata and generate tags.

3. Implement Scene Tagging and Indexing

Use ML models to automatically tag scenes (e.g., pedestrian, traffic sign) and store tags in a searchable index (e.g., Elasticsearch). Also support manual tagging and corrections. Index both metadata and references to raw data.

4. Design Query API and Storage Tiering

Provide a query API that allows engineers to search by tags, time ranges, and other metadata. Implement storage tiering: hot storage (SSD) for recent/frequently accessed data, warm (HDD) for less frequent, and cold (object storage like S3 Glacier) for archival. Use lifecycle policies to move data automatically.

5. Address Access Control and Security

Implement role-based access control (RBAC) to restrict data access based on user roles. Ensure encryption at rest and in transit, and audit logging for compliance.

Key Points to Mention

  • Use of distributed message queue (e.g., Kafka) for ingestion to handle high throughput and decouple components.
  • Automated tagging using ML models (e.g., object detection) and manual tagging for ground truth.
  • Indexing strategy: inverted index (Elasticsearch) for metadata and tags, with references to raw data in object storage.
  • Storage tiering: hot/warm/cold with lifecycle policies to optimize cost and performance.
  • Query API design: support complex queries (e.g., boolean, time-range) and return results with links to raw data.
  • Access control: RBAC, encryption, and audit logs to ensure data security and compliance.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.