Implementing Machine Learning at Scale in Elixir using Bumblebee and Broadway

In this talk recorded at Code BEAM Europe 2023, Raj Rajhans explores how organizations can effectively implement machine learning at scale within the Elixir ecosystem. He demonstrates how the Elixir libraries Bumblebee and Broadway can tackle machine learning tasks in production, using the example of natural language media search. Raj showcases the implementation steps, including the use of CLIP models from OpenAI and leveraging the Bumblebee library to integrate models from Hugging Face into Elixir applications. He explains how to use Broadway for batch processing and indexing bulk media using Bumblebee and Nx, leading to efficient and scalable machine learning solutions. The talk also delves into storing embeddings in vector databases and configuring Broadway for high concurrency and throughput. Raj concludes by demonstrating a proof-of-concept and discusses the challenges faced and solutions found in the process.