Back office ops · Production

How Dropbox Dash built scalable multimedia search with just-in-time previews and metadata-first indexing

The problem

Knowledge workers routinely lose time finding images, videos, and audio files across apps because media files often have cryptic default names, lack meaningful metadata, and require significantly more compute to process and rank than text documents.

Workflow diagram · grounded in source
1
Media metadata indexed at ingestion
integration
“For example, we extract features such as file path, title, and EXIF. These metadata provide a lightweight foundation that enables basic search functionality with minimal processing overhead.”
2
User initiates media search
trigger
“When a user searches for media, we configure the query to match their input against the metadata features extracted during indexing”
3
Multi-phase retrieval and ranking
ai_action
“Dash operates a multi-phase retrieval and ranking scheme, which was previously trained and optimized for textual content. Retrieving and ranking multimedia content requires having indexed any new multimedia-specific signals”
4
Geolocation-aware query processing
ai_action
“we index a GPS location as a chain of IDs corresponding to the geographical hierarchy. For instance, we can look up the GPS coordinates of a photo to be from San Francisco in a process known as "reverse geocoding."”
5
Just-in-time preview generation
output
“To power the just-in-time approach, we rely on an internal previews service built on top of Riviera, a framework originally developed for Dropbox Search.”
6
Results and metadata rendered
output
“When users request more detail—such as camera metadata or timestamp—we fetch it on-demand via a separate endpoint. This keeps the initial search response lean while still supporting deeper inspection when needed.”
Reported outcome

Dropbox Dash delivered a robust multimedia search experience with metadata-first indexing, geolocation-aware queries, and just-in-time previews, substantially reducing latency while ingesting approximately 97% of media files.

Reported metrics
Image file size vs non-media files3X larger
Video file size vs non-media files13X larger
Media files ingested97%
Preview cache duration30 days
Show all 5 reported metrics
image file size vs non-media files3X larger
video file size vs non-media files13X larger
media files ingested97%
preview cache duration30 days
system latency improvementsubstantially reduce latency
Reported stack
Dropbox DashRivieraCanva
Source
https://dropbox.tech/infrastructure/multimedia-search-dropbox-dash-evolution
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Dropbox Dash delivered a robust multimedia search experience with metadata-first indexing, geolocation-aware queries, and just-in-time previews, substantially reducing latency while ingesting approximately 97% of medi…

What tools did this team use?

Dropbox Dash, Riviera, Canva.

What results were reported?

Image file size vs non-media files: 3X larger; Video file size vs non-media files: 13X larger; Media files ingested: 97%; Preview cache duration: 30 days (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Media metadata indexed at ingestion → User initiates media search → Multi-phase retrieval and ranking → Geolocation-aware query processing → Just-in-time preview generation → Results and metadata rendered.

Related back office ops cases