Chris Bailey

In 2026, the “Thinking Gap” is widening. As LLMs become the default shortcut for assignments, the traditional “homework” model has collapsed into an arms race of prompt engineering. How do we reclaim human critical thinking? Enter ForgeThought: a local-first, “Sovereign Classroom Appliance” designed to move cognitive synthesis back into the physical classroom.

This talk explores the architectural journey of building an air-gapped learning node on the BEAM. We’ll dive into how Phoenix LiveView orchestrates real-time “Cognitive Sprints” and how Nx and Bumblebee allow us to run high-performance AI inference on the edge, verifying student logic against a teacher’s secret “Context Keys” without ever touching the cloud.

The Highlight: A live, interactive demo where the audience joins as “students.” We will attempt to host up to 30 concurrent participants on a local workstation, demonstrating how to manage high-latency AI validation queues with Oban and sync stateful UI transitions across a “noisy” network. We’ll discuss the “Return to the Monolith,” the ethics of private AI infrastructure, and why Elixir is uniquely positioned to save education from the “Black Box” of the cloud.

Key takeaways:

In 2026, the “Thinking Gap” is widening. As LLMs become the default shortcut for assignments, the traditional “homework” model has collapsed into an arms race of prompt engineering. How do we reclaim human critical thinking? Enter ForgeThought: a local-first, “Sovereign Classroom Appliance” designed to move cognitive synthesis back into the physical classroom.

This talk explores the architectural journey of building an air-gapped learning node on the BEAM. We’ll dive into how Phoenix LiveView orchestrates real-time “Cognitive Sprints” and how Nx and Bumblebee allow us to run high-performance AI inference on the edge, verifying student logic against a teacher’s secret “Context Keys” without ever touching the cloud.

The Highlight: A live, interactive demo where the audience joins as “students.” We will attempt to host up to 30 concurrent participants on a local workstation, demonstrating how to manage high-latency AI validation queues with Oban and sync stateful UI transitions across a “noisy” network. We’ll discuss the “Return to the Monolith,” the ethics of private AI infrastructure, and why Elixir is uniquely positioned to save education from the “Black Box” of the cloud.

1. Local-First AI on the Edge: Practical strategies for deploying and optimizing quantized LLMs using Bumblebee and Nx.
2. Handling Spiky Inference: How to use Oban and GenServers to queue and throttle CPU-intensive AI tasks without breaking the “snappy” LiveView UX.
3. The Sovereign Tech Stack: Why the BEAM’s fault tolerance and resource management are ideal for “Appliance-based” hardware in unpredictable environments.
4. Social Impact and Ethics: A framework for building “Anti-Cheating” tools that prioritize student privacy and “Neural Weightlifting” over surveillance. How engineers can use their skills to solve the “Cognitive Offloading” crisis in schools.

Target audience:

• Senior developers, software architects, and EdTech enthusiasts interested in Local-First software, Machine Learning on the BEAM, and mission-driven engineering.