Episode

Nx and Machine Learning in Elixir with Sean Moriarity

Podcast: Elixir Wizards
Published: Jun 19, 2025
Duration seconds: 2661
Processing state: processed
Canonical source: https://smartlogic.fireside.fm/s14-e04-nx-machine-learning-elixir-sean-moriarity
Audio: https://aphid.fireside.fm/d/1437767933/03a50f66-dc5e-4da4-ab6e-31895b6d4c9e/53f845b4-fada-46fc-ada0-0449ce84fb6a.mp3
JSON: /v1/public/podcasts/elixir-wizards/episodes/nx-and-machine-learning-in-elixir-with-sean-moriarity
Markdown: /podcast/elixir-wizards/nx-and-machine-learning-in-elixir-with-sean-moriarity.md

Actions

POST https://stenobird.com/v1/public/podcasts/elixir-wizards/episodes/nx-and-machine-learning-in-elixir-with-sean-moriarity/transcription-requests
Idempotently request low-priority transcript generation for this episode.
GET https://stenobird.com/podcast/elixir-wizards/nx-and-machine-learning-in-elixir-with-sean-moriarity.md
Read the agent-friendly Markdown representation of this episode resource.

Summary

Sean Moriarity discusses the evolution of the Nx ecosystem from native model implementation to a powerful orchestration layer for LLMs. He shares strategies for integrating Elixir with Python-based ML libraries and leveraging Elixir's distributed computing for scalable AI workloads.

Topics

Machine Learning
Elixir
Nx
LLMs
Distributed Computing
Python Interoperability
Numerical Computing
Software Engineering

Highlights

Main idea: The Elixir ML landscape is shifting from building native models to using Elixir as a high-level orchestration layer for external tools
Practical takeaway: Use Elixir's distributed capabilities to manage complex ML workflows and coordinate between different computing nodes
Strategy: When introducing Elixir to ML teams, frame its strengths in terms of familiar concepts like Python's Ray framework for distributed computing
Failure mode: Avoid the political challenge of trying to replace established Python ecosystems; instead, focus on bridging the two via interoperability
Practical takeaway: Leverage libraries like Instructor for structured outputs and Bumblebee for running pre-trained models within the Elixir ecosystem

Chapters

1:00 Career Updates and the Future of ML in Elixir: Sean discusses his transition to Magic.dev and the recent advancements in the Nx and Bumblebee ecosystems.
4:20 Elixir as an Orchestration Layer: Exploring how Elixir's strengths in orchestration complement numerical computing and model management.
7:35 Structured Outputs and Ecosystem Tools: A look at powerful libraries like Instructor and the impact of modern LLM capabilities on Elixir development.
10:55 The Boundaries of Nx and Axon: Defining the roles of Nx as a numerical framework and Axon as a higher-level neural network library.
14:20 Native ML vs. Hybrid Approaches: Discussing the trend of blending native Elixir ML implementations with vector search and embedding workflows.
17:35 The Accessibility of Modern Machine Learning: How the rise of high-level frameworks has lowered the barrier to entry for applying ML to real-world problems.
20:50 Navigating Organizational Change: The social and political challenges of introducing Elixir into established Python-centric machine learning organizations.