Design street-view image ingestion and storage system

Prompt

Design a system to ingest, store, and process street-level images for a "Street View"-like product.

Scenario

• A fleet of taxis is equipped with cameras.
• Each taxi periodically captures photos (optionally with sensor metadata like GPS, timestamp, heading) and uploads them to your backend over unreliable mobile networks.
• Stored images are consumed by downstream systems for:
  • Image understanding / feature extraction
  • Map generation (stitching/selection)
  • Display to end users (serving tiles/panoramas or individual frames)

What to cover

1. Requirements gathering
   • Functional requirements (ingestion, storage, retrieval, processing hooks)
   • Non-functional requirements (durability, availability, latency, cost, privacy/security)
   • Scale assumptions (taxis, images per taxi, image size, retention)
2. High-level architecture
   • Main services and data stores
   • Data flow from taxi → ingestion → durable storage → processing pipeline → serving
3. Upload API design
   • How the taxi uploads images (single-shot vs chunked/resumable)
   • Response semantics (sync/async), idempotency, retries
   • Handling bad networks, offline buffering, and partial uploads
4. Security / authentication
   • How devices authenticate (tokens/certs)
   • How to prevent abuse and secure data in transit/at rest
   • What to do if a token/device credential is compromised (revocation, rotation, blast radius reduction)
5. Trade-offs & extensions
   • Storage choices and partitioning
   • Exactly-once vs at-least-once ingestion
   • If given more time, what additional features or safeguards you would add (and why)