D
DreamLake

DB Schema

Schema Overview

BSS uses a minimal MongoDB database for indexing only. All binary data and metadata lives in S3. The DB records are pointers for querying — not the source of truth for content.

Models

ModelDescription
VideoVideo index record (fps, length, duration, timestamps)
AudioAudio index record (sampleRate, channels, codec, duration)
TextTrackText track index (entryCount, time range, format)
LabelTrackLabel track index (entryCount, time range)
EmbeddingVector embeddings for semantic search

Key Design Principles

  • S3 is the source of truth — DB records are indexes, not data stores
  • IDs are MongoDB ObjectId — auto-generated
  • Soft deletesdeletedAt on Video, Audio, TextTrack, LabelTrack
  • Owner/project scoping — all records carry owner + project for namespace isolation
  • embId — optional link to a VectorIndex for semantic search (Audio, TextTrack, LabelTrack)