Back office ops · Production

Dropbox integrates Mobius Labs' Aana multimodal AI models into Dash for scalable media understanding

The problem

Content spanning text, images, audio, and video is scattered across countless apps and tools, making it hard to search and find insights quickly—and processing that content at exabyte scale becomes cost-prohibitive with conventional architectures.

Workflow diagram · grounded in source
1
Multimodal content ingestion
trigger
“It takes in files of all kinds—demo videos, audio interviews, photo libraries—and analyzes them together”
2
Audio modality processing
ai_action
“For audio, it uses inference-optimized Whisper-like models developed with open-source collaborators, such as the faster-whisper-large-v3-turbo model”
3
Vision and language processing
ai_action
“Its vision and language systems rely on transformer-based and mixture-of-experts (MoE) architectures engineered for fast, cost-effective inference on off-the-shelf GPUs”
4
Cross-modal relationship analysis
ai_action
“Unlike systems that treat text, images, audio, and video as separate streams, Aana looks at how they relate to one another, revealing patterns and insights that emerge only when these modalities are combined”
5
Shared vector space indexing
output
“All of this information is distilled into a shared vector space, enabling fast, multimodal search”
6
Natural language search
output
“You can ask for 'the part where the presenter explains the API flow' instead of scrubbing through timestamps or relying on basic metadata”
Reported outcome

Aana enables Dropbox Dash to analyze multimedia content at exabyte scale with dramatically lower compute costs than conventional architectures, enabling natural language queries across video, audio, and image content without manual searching.

Reported metrics
Computational requirements vs conventional architecturessignificantly lower computational requirements
Compute and memory costsdramatically lower compute and memory costs
Compute footprint vs traditional architecturesa fraction of the compute footprint
Reported stack
AanaAana SDKHQQGemliteDropbox Dashfaster-whisper-large-v3-turbo
Source
https://dropbox.tech/machine-learning/mobius-labs-aana-dropbox-multimodal-understanding
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Aana enables Dropbox Dash to analyze multimedia content at exabyte scale with dramatically lower compute costs than conventional architectures, enabling natural language queries across video, audio, and image content…

What tools did this team use?

Aana, Aana SDK, HQQ, Gemlite, Dropbox Dash, faster-whisper-large-v3-turbo.

What results were reported?

Computational requirements vs conventional architectures: significantly lower computational requirements; Compute and memory costs: dramatically lower compute and memory costs; Compute footprint vs traditional architectures: a fraction of the compute footprint (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Multimodal content ingestion → Audio modality processing → Vision and language processing → Cross-modal relationship analysis → Shared vector space indexing → Natural language search.

Related back office ops cases