R. Riley Holmes¶
I turn complex system problems into working software, fast.
Currently writing Elixir for satellite ground control systems at True Anomaly.
I’m a software developer with a safety-critical systems background. I help teams tackle the hard infrastructure problems where data models are tangled, subsystems are evolving, and real-world conditions make reliability difficult.
Have a hard problem? Let's talk →
GitHub ·
LinkedIn
Areas I Work In¶
- AI Agent Reliability & Orchestration: Safety-critical engineering principles applied to a space that desperately needs them.
- Data Integration & Model-Based Architecture: Formal ontology management, schema evolution, and fault-tolerant integration.
- Rapid Prototyping & System Delivery: Complex systems, built and deployed quickly. I specialize in going from “we need this” to production without the months of overhead.
Featured Projects¶
An Elixir library for dynamic models and metamodels, with transformations and arbitrary levels of abstraction.
HexDocs · Live demo
Nix infrastructure for Phoenix apps: a flake template that pins dependencies, exposes a dev shell, builds a release, and deploys.
GitHub · Write-up
A persistent, autonomous LLM assistant built as a thin (~200-line) Nix systemd module that slots into a NixOS server.
GitHub · Write-up
The research foundations behind Paradigm's approach to runtime schema management.
Read the PDF
See all projects →
Recent Blog Posts¶
Introducing Groundhog: a persistent, autonomous LLM assistant built as a thin ~200-line Nix systemd module that slots into a NixOS VPS.
Evolving a NixOS deploy-rs setup where each site is a flake input, the whole server builds atomically, and multi-node deploys run with one command.
Overhauling a home network of cheap ProDesk mini-PCs into a NixOS fleet deployed over the network with deploy-rs.
NixOS' QEMU-based integration tests spin up whole multi-node systems in CI—why that makes them even more undeniable than container-based testing.
The agent harness—the control plane between your code and the model—is the next bottleneck. Why cramming it into your context is a costly mistake.
Threads behind the most successful AI coding agents—Claude Code, Gas Town, OpenClaw—and a workable path forward for agent architecture.
A walk through three progressively stronger tiers of decoupling in Elixir—map conventions, behaviours, and protocols—drawn from a graph-schema library.
Cutting through the 'Agent' hype: why Elixir/OTP already gives you fault-tolerant LLM pipelines on the BEAM without a bloated framework.
An approachable explainer untangling the four things all called 'Nix'—the language, the tool, the nixpkgs repo, and the NixOS operating system.
Why I'm long-term bearish on Docker, Kubernetes, and managed clouds—and how NixOS solves dependencies and deployment from first principles.