Elixir Mentor Podcast: Forecasting, AI Agents, and the BEAM

I joined the Elixir Mentor podcast to talk about Soothsayer, why the BEAM is the best runtime for AI workflows, and agentic commerce.

Podcast microphone in a studio
Photo by Will Francis on Unsplash

I joined Jacob Lutzko on the Elixir Mentor podcast to talk about Soothsayer, time series forecasting on the BEAM, and why Elixir is quietly becoming the best ecosystem for AI and agentic workflows.

We covered a lot of ground:

  • Soothsayer: how I built a time series forecasting library inspired by NeuralProphet, using Axon and Nx. It started as a tool for predicting engineering metrics at SourceLevel and turned into an open source project. The cool part: because Axon is so high-level, the codebase is dramatically more readable than the original Python implementation.
  • Why Elixir is great for AI backends: agent calls are long-running HTTP requests. Most ecosystems struggle with that kind of concurrency. The BEAM was built for it. Python teams building production AI workflows are already looking elsewhere, usually to TypeScript, but the BEAM is a much better fit.
  • Agent frameworks reinventing OTP: Python AI frameworks are rediscovering message passing, actor models, and fault tolerance. We've had all of that battle-tested for decades.
  • Agentic commerce at New Generation: what I'm working on now. Conversational search replacing keyword search, agents browsing e-commerce sites on behalf of users, and the rise of AEO (Agent Engine Optimization) as the next SEO.
  • Running ML models inside your Elixir app: with Bumblebee and Nx, you can serve neural networks as processes in your supervision tree. No separate model serving infrastructure. No extra deployment. Just call it like any other module.
  • A RAG system for Elixir: an open source idea I've been thinking about. Embeddings and vector search directly in Elixir, without needing a separate vector database. For most use cases (customer service bots, FAQ systems), the data fits in RAM and the BEAM handles the parallelism.
  • Livebook in the wild: I met journalists at the San Francisco meetup who use Livebook instead of Jupyter notebooks for data analysis. They prefer it.
  • Learning Ash with LLMs: when I joined New Generation, I had never used Ash. I learned it by having conversations with Claude Code, asking it to explain policies, actions, and why things are structured the way they are. Surprisingly effective.

We also went on tangents about connecting Livebook to production nodes (powerful and terrifying), Jose Valim writing Devise and then deliberately not recreating it for Elixir, password hygiene, mechanical keyboards, and floppy disks.

Thanks to Jacob for having me on. If you're into Elixir, subscribe to Elixir Mentor.


Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to George Guimarães..

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.