Universal Content Processing Platform

7/20/2026

The post details the evolution of Dropbox's Riviera platform from a file preview service to a universal content processing platform. It highlights the architectural shift from product-specific pipelines to reusable transformations, enabling support for over 300 file formats and diverse workloads. The platform's plugin architecture allows for independent scaling and maintenance of transformation capabilities. The integration with AI products like Dash is emphasized, showcasing how Riviera prepares content for AI models. The post also announces the external availability of Riviera's capabilities through APIs and Model Context Protocol.

6/25/2026

This post details the application of DSPy, an AI optimization framework, to improve the Dash chat agent. It describes the process of calibrating LLM judges against human-labeled examples to ensure their evaluations align with human judgment. The post explains how DSPy's optimization algorithms (GEPA, MIPROv2) were used to refine judge prompts and subsequently optimize the chat agent's system prompt. This created a feedback loop where human labels improved judges, judges provided evaluation signals, and those signals enhanced the agent, resulting in fewer incomplete answers and reduced token usage.

2/12/2026

This post details how Dropbox leverages low-bit inference techniques, specifically quantization, to improve the efficiency and reduce the cost of running large AI models for products like Dropbox Dash. It explains the trade-offs between different quantization formats (e.g., 16-bit to 4-bit), the role of hardware acceleration (Tensor Cores, Matrix Cores), and the challenges of bitpacking for sub-byte formats. The post highlights how these optimizations are crucial for delivering responsive user experiences and managing resource usage in production AI workloads.