Posts on JoeSindel.com

One-to-Many Beacon Nodes: Rethinking PoS Validator Topology

Sun, 10 May 2026 00:00:00 +0000

A proof-of-stake validator is not a monolith. It’s three processes coordinating tightly: a validator client signing attestations and proposing blocks, a beacon node participating in consensus, and an execution client running EVM state. The simplest deployment binds them 1:1:1 — one validator paired with one beacon, one beacon paired with one execution client. That topology is fine when you’re running 32 ETH worth of stake on a home setup. It is not fine when you’re operating at the scale of a top-tier staking provider.

What Major Incident Command Actually Buys You

Sun, 10 May 2026 00:00:00 +0000

A SEV at company scale isn’t a bigger version of an outage on a small team. It’s a different problem. Once you have dozens of services on the page, multiple product surfaces affected, regulators watching, executives asking questions, and customer comms running in parallel with engineering response — the thing that breaks first is coordination, not engineering capability.

This post is about what the IMAG / Incident Command System framework actually buys you in that environment, where I’ve seen it get misapplied, and the part most teams skip: making the post-mortem loop close.

Building a Private AI Server with NVIDIA Jetson AGX Thor

Sat, 25 Apr 2026 00:00:00 +0000

A few days ago I got my hands on an NVIDIA Jetson AGX Thor developer kit. 128GB of unified memory, a Blackwell GPU, and enough raw compute to run serious language models locally. This post covers the full build: three inference backends, voice chat from my phone over cellular, video-based object detection, a complete monitoring pipeline, and a custom dashboard. Everything survives a reboot, nothing touches the cloud.

The Hardware

The Jetson AGX Thor is not a consumer product. It’s a developer kit built for robotics and edge AI workloads. The specs that matter for LLM inference: