AI-Powered Search at Scale - Jeff Weiss | ElixirConf EU

[CodeSync]

AI-Powered Search at Scale - Jeff Weiss | ElixirConf EU 2026

AI-Powered Search at Scale - Jeff Weiss | ElixirConf EU 2026

This talk was recorded at ElixirConf EU 2026. Check our upcoming events at https://elixirconf.eu/

---

At Frame.io (Adobe), AI-powered search helps creative professionals locate specific shots across millions of assets using natural language and semantic visual search. This session explores the architecture behind integrating Llama 4 via AWS Bedrock to translate complex queries into OpenSearch DSL, alongside the use of CLIP-based models for vector embeddings.

Jeff Weiss discusses the architectural hurdles of processing video embeddings, which can take over 30 seconds per asset. You will learn why the team moved beyond their existing Oban infrastructure to a multi-stage Broadway pipeline consuming from SQS. This approach allowed for independent scaling, traffic buffering, and complete isolation from critical background jobs. Jeff shares practical patterns for LLM integration, strategies for choosing between Broadway and Oban, and how to build highly observable AI pipelines within the Elixir ecosystem.

---

Comments welcome! View the code-sync and elixirconf-eu tags for more Code Sync and ElixirConf EU talks!