In this episode of the Elixir Mentor Podcast, I chat with George Guimarães about Soothsayer, his time series forecasting library inspired by NeuralProphet and built on Axon and NX for business data analysis.

We explore how Soothsayer decomposes business data into seasonal components, handles holidays as special events, and uses neural networks to model nonlinear patterns. George explains why Elixir's ecosystem with NX, Axon, and Bumblebee provides unique advantages for machine learning workflows, allowing you to run models directly in your supervision tree without external infrastructure.

The conversation expands into why Elixir is particularly well-suited for AI agent development. George shares insights from his current work building agentic commerce solutions, where the BEAM's actor model, fault tolerance, and message passing provide battle-tested patterns that other ecosystems are now trying to replicate for LLM workflows. We also discuss AEO (Agent Engine Optimization) as the new SEO, and how websites will evolve to serve both human and agent visitors.

Whether you're interested in time series forecasting, building AI-powered applications in Elixir, or understanding why the BEAM's concurrency model is perfect for the agentic future, this conversation offers valuable perspective from an Elixir community OG.

Resources Mentioned:
- Soothsayer: https://github.com/georgeguimaraes/soothsayer

Connect with George:
- X: https://x.com/georgeguimaraes
- Website: https://georgeguimaraes.com

SUPPORT ELIXIR MENTOR
- Elixir Mentor: https://elixirmentor.com/?utm_source=elixir-mentor

In this episode of the Elixir Mentor Podcast, I chat with Isaak Tsalicoglou, author of Elixir Software Engineering. Isaak shares hard-won lessons from building production Elixir API clients, covering validation strategies, error handling approaches, authentication flows, and architectural patterns that actually work in the real world.

We explore Isaak's journey back into programming through building internal tools for his family's industrial equipment business, and how that led to writing a comprehensive guide on REST API client development. He explains his approach to request validation, why he ultimately decided against using Ecto schemas for API responses, and the importance of resisting unnecessary complexity in software architecture.

The conversation covers practical API design topics including how to structure clean RESTful routes, avoiding tight coupling between APIs and UIs, and finding the right balance between over-serving and under-serving data. Isaak also shares his thoughts on LLM-assisted development, explaining why he prefers using AI as a code reviewer rather than fully automated coding, and discusses his self-hosting infrastructure setup for privacy-conscious applications.

This episode offers valuable insights for anyone building API clients in Elixir or thinking critically about software architecture decisions and their long-term implications.

Resources Mentioned:
- Elixir Software Engineering: https://leanpub.com/elixir-software-engineering

Connect with Isaak:
- X: https://x.com/realMrLaminar
- LinkTree: https://linktr.ee/tisaakx

SUPPORT ELIXIR MENTOR
- Elixir Mentor: https://elixirmentor.com/?utm_source=elixir-mentor

In this episode of the Elixir Mentor Podcast, I chat with Elixir creator José Valim about Tidewave, the AI coding agent that lives inside your web framework. José shares the journey from falling in love with MCP to discovering its limitations, and how Tidewave solves the copy-paste problem that plagues AI-assisted development.

We explore how Tidewave integrates directly with your browser to eliminate tedious workflows, automatically detecting exceptions and validating changes without manual intervention. José explains why the tool now supports Phoenix, Rails, Django, FastAPI, Flask, and Next.js, and how building one feature benefits all frameworks simultaneously. The conversation covers prompting strategies, context management, and the unique challenges of building Tidewave with Tidewave.

José offers candid insights on why MCP has fundamental limitations for user experience, the security concerns around AI agents, and why Elixir's message-passing architecture makes it ideal for building agentic systems. We discuss the evolving type system in Elixir, code review workflows with AI, and the upcoming Tidewave features including multi-element inspection and symbol search.

The episode concludes with José's perspective on transitioning from open source maintainer to product owner, collecting user feedback through Discord, and exciting developments with Tauri for building desktop applications with Elixir. This conversation provides valuable insights for developers interested in AI-powered tooling and the future of web development.

Resources Mentioned:
- Code Benchmark: https://github.com/Tencent-Hunyuan/AutoCodeBenchmark

Connect with José:
- X: https://x.com/josevalim
- Tidewave: https://tidewave.ai
- Dashbit: https://dashbit.co

SUPPORT ELIXIR MENTOR
- Elixir Mentor: https://elixirmentor.com