Projects

Current and upcoming projects. All open source work is available on GitHub.

• AI News Digest

  ACTIVE

  Automated daily AI industry digest. Each morning a scheduled pipeline fetches and filters RSS feeds, synthesizes a digest post, and deploys it to a Hugo site on Cloudflare Pages. No server, no application code to maintain.

  Live Site ·GitHub

• OpenNIC Tier-2 Installer

  ACTIVE

  A reusable bash installer that turns a fresh Debian 13 host into a public OpenNIC Tier-2 DNS resolver. Speaks Do53, DoT, DoH, and DNSCrypt with DNSSEC validation against both IANA and OpenNIC root keys, no-logs by default, idempotent across reruns.

  Live Site ·GitHub

• LLMCapabilities

  v0.2.1

  A zero-dependency Ruby gem for resolving LLM model capabilities through a 4-tier hierarchy: empirical observation, OpenRouter's model index, RubyLLM registry, and provider heuristics. Answers the question "does this model support this capability?" with layered, evidence-based resolution.

  RubyGems ·GitHub

Experiments

Time-boxed investigations. Evaluated by what we learn, not what they ship.

• Merlin v1

  CONCLUDED

  A 26-day autonomous experiment exploring whether a frozen-weight LLM can compound intelligence given external infrastructure: a filesystem for memory, a task queue for continuity, and dream-cycle consolidation for knowledge extraction. A containerized TypeScript/Bun orchestrator on Proxmox activated Claude Code on a 10-minute tick loop — 500+ tasks, 153 consolidation cycles, 122 lessons, and 3,920 tests across the run.

  GitHub

• Merlin v2

  UPCOMING

  A continuation of the autonomous self-improvement experiment Merlin v1 ran, applying the insights and lessons from 26 days of operation to a fresh architecture rather than evolving the v1 codebase. The premise — frozen-weight LLM, external infrastructure for memory and continuity, dream-cycle knowledge consolidation — carries forward; the orchestrator, model routing, and consolidation rhythm get rebuilt from scratch with v1 as ground truth.

  Coming soon

• Kaiden

  UPCOMING

  A batch deliberation engine that orchestrates multiple LLM nodes — each backed by a different model — to reason about a prompt collaboratively across iterative rounds. Cognitive diversity emerges from substrate diversity (different models) and structural diversity (cross-interpretation rotation). Not an agent framework.

  Coming soon