Dropbox Evolves Riviera Content Processing Platform to Support AI Workloads

Dropbox has transformed its Riviera content processing platform from a file preview service into a comprehensive platform supporting over 300 file formats and 100 transformation capabilities. The platform now handles hundreds of thousands of transformations per second, integrating Search, Replay, Sign, and Dash to support AI workloads.

Initially developed internally for generating previews of Dropbox's various file formats, engineers restructured content processing into reusable transformations, enabling the creation of diverse pipelines. The platform separates orchestration from execution, with a central component managing request validation, pipeline composition, worker dispatch, and caching.

Individual workers execute specific transformations, while a plugin model allows for the addition of new capabilities without altering the core. Riviera's architecture has already been utilized in AI processing, converting content into text and embeddings for AI-powered summaries and question answering. Dropbox has expanded its use of the platform to Dash, a product requiring content preparation for search and AI operations.

The company has also made Riviera capabilities accessible through public APIs, offering asynchronous endpoints for converting documents to Markdown, transcribing audio and video, and extracting structured metadata. These Markdown API endpoints support applications such as indexing, rendering, and feeding documents into large language models.

The platform's integration with external developer tools, like conductor-dropbox, an open-source integration for Conductor OSS and Orkes, exemplifies its adaptability in document and AI processing workflows. Dropbox's Riviera stands apart from general-purpose extraction frameworks like Apache Tika, which primarily offer APIs for detecting, parsing, and extracting text and metadata from various file formats.