Elixir Mentor George Guimarães on Forecasting
10 snips
Dec 21, 2025 Join Brazilian software engineer George Guimarães, the mind behind Soothsayer, as he dives into the world of time-series forecasting using Elixir. He discusses how Soothsayer tackles seasonal trends and holiday impacts with neural networks. George shares why Elixir's ecosystem excels in AI development, highlighting the benefits of its concurrency model. He even introduces the concept of Agent Engine Optimization (AEO) as the next big thing for enhancing user experience. Expect insightful takes on AI, agentic commerce, and the Elixir community!
AI Snips
Chapters
Transcript
Episode notes
Soothsayer Born From A Failed Startup Pivot
- George started Soothsayer after trying forecasting in his metrics startup and wanting better Elixir-native tools.
- He ported ideas from NeuralProphet into an Axon/NX implementation and open-sourced it when the startup didn't continue.
Decompose Business Data For Clear Forecasts
- Soothsayer decomposes business time series into trend, seasonalities, holidays, and other components for explainable forecasting.
- George built it on Axon and NX to leverage neural networks for faster convergence and clearer model structure.
Use Neural Nets For Nonlinear Business Patterns
- Use neural networks to model nonlinear regressors and lagged features when forecasting business metrics.
- Accept a bit less explainability for faster convergence and richer nonlinear pattern capture.