Robert Allen is a systems/platform engineer, open source maintainer, and sheep and poultry farmer in Farmville, Virginia. By day, he drives infrastructure automation for a major education technology company. By evening, he builds developer tools that solve real problems in technology and agriculture — and somehow also runs a regenerative pasture farm.
- AI Memory Systems — Building structured, persistent memory for AI agents: subcog and mnemonic
- Open Standards for AI Tooling — Authoring the MIF specification to make AI memory portable and interoperable
- Agentic Workflows — GitHub Agentic Workflows (
gh-aw) for autonomous, event-driven repository operations at scale - Supply-Chain Security — Signed, SLSA-attested, fail-closed-verified release pipelines (cosign, SBOM, provenance) as reusable org workflows
- AI-Assisted Development — Claude Code plugins, agents, and multi-agent orchestration
- DevOps & Platform Engineering — Infrastructure automation, CI/CD, AWS architecture
Memory Interchange Format (MIF) — Portable AI Memory
The AI memory ecosystem is fragmented — Mem0, Zep, Letta, and others all use proprietary schemas with no interoperability. MIF defines a common data model with dual representations: human-readable Markdown (Obsidian-compatible) and machine-processable JSON-LD, with lossless conversion between them. Three conformance levels scale from a 4-field core to full bi-temporal provenance, decay, and embeddings, with migration guides from the major memory systems.
Status: v0.1.0-draft • Specification • GitHub
The intersection of cognitive science and AI systems presents a compelling question: how do we build AI that remembers meaningfully? Human memory isn't a tape recorder — it's a constructive process where our mental models (ontology) shape what we encode, and our memories reshape how we understand the world.
MIF grew directly from this research, the product of five-plus iterations on memory systems — subcog and mnemonic are its public expressions. Alongside them, I'm writing an academic paper measuring memory system impact across LLM models; one early finding is how much variance exists in structured recall between models in ways general benchmarks don't predict.
subcog — Persistent Memory for AI Coding Assistants
A Rust memory system that captures decisions, learnings, and context from coding sessions. Hybrid search (semantic + BM25), MCP server integration, SQLite persistence with a knowledge graph, and proactive memory surfacing. Its filesystem-native sibling, mnemonic, is a pure MIF Level 3 implementation — no dependencies, just markdown files and git.
rlm-rs — Recursive Language Model CLI
A Rust CLI implementing the RLM pattern (arXiv:2512.24601) for Claude Code — process documents 100x larger than the context window through intelligent chunking, SQLite persistence, and recursive sub-LLM orchestration. Ships with a companion plugin.
nsip — Sheep Genetic Evaluation CLI & MCP Server
A Rust CLI and MCP server for the National Sheep Improvement Program database — 400,000+ animals with EBVs, pedigrees, and performance data. Provides breeding intelligence: Wright's inbreeding coefficients, weighted trait ranking, mating recommendations, and flock summaries. nsip-example demonstrates GitHub4Farms — GitHub Issues as a farm record-keeping system with Copilot-powered genetic enrichment, accessible to farmers with no technical background.
This organization's .github repo centralizes signed, SLSA Build L3-attested, fail-closed-verified container delivery: keyless cosign signatures, CycloneDX SBOMs, provenance attestations as OCI referrers, verification gates before any publish or deploy, and change-record-gated production promotion — all consumable by any repo as reusable workflows. It also runs 17 gh-aw agentic workflows for daily triage, standup, dependency housekeeping, and retrospectives across the org.
Last updated: 2026-06-11
Ranked by recent contributions, community engagement, and development activity.
| Repository | Description | Tech | Activity |
|---|---|---|---|
| .github | Organization shared workflows, policies, and community healt... | Shell | ✨ Active |
| nsip | Sheep genetic evaluation CLI & MCP server -- search animals,... | Rust | ✨ Active |
| rlm-rs | Rust CLI implementing the Recursive Language Model (RLM) pat... | Rust | 📈 Growing |
| adr | Claude Code plugin for complete ADR lifecycle management wit... | MDX | 📈 Growing |
| subcog | Persistent memory system for AI coding assistants. Captures ... | Rust | 📈 Growing |
| Rhubarb | A PHP Celery Client Implementation | PHP | 📈 Growing |
| structured-madr | Machine-readable Architectural Decision Records with YAML fr... | MDX | 📈 Growing |
| mnemonic | Persistent filesystem-based memory system for Claude Code. P... | Python | 📈 Growing |
- copilot-review (JavaScript) - A Claude Code plugin that uses GitHub Copilot as the backend for code reviews an...
- fastmcp-lro (Python) - Large Result Offloading for MCP servers. Writes oversized tool responses to JSON...
- autoresearch (Python) - Autonomous skill improvement loop for Claude Code plugins — inspired by Karpathy...
Languages Rust | Python | TypeScript | Go
Infrastructure AWS | Docker | Kubernetes | Terraform | GitHub Actions
AI Integration Claude Code | GitHub Copilot | MCP Protocol
Supply Chain cosign | SLSA | CycloneDX | OCI referrers
Specifications MIF (mif-spec.dev)
Platforms Linux | macOS | AWS (Solutions Architect certified)
- Developer Experience First - Tools should reduce friction, not add it
- Automation Over Documentation - Encode knowledge in code
- Open by Default - Share solutions that might help others
- Practical Over Perfect - Ship working software, iterate based on usage
- GitHub Issues - For project-specific discussions
- Pull Requests - The best way to propose changes
- LinkedIn - Professional networking and collaboration