About
What I build: infrastructure that makes AI coding agents actually do work. Document retrieval (alcove), paper research (research-agent), agent execution harness (epic-harness), book publishing (Velith) — each tool runs as both a CLI and an MCP server, and installs as a plugin on five or more agent hosts.
How I build: single Rust binaries, SQLite with FTS5, static sites. I prefer things that deploy with no build step and no backend. Research runs as a trackable pipeline — collect, gap-analyze, fact-check, publish — and its outputs are the Research Sites below.
Before this: commerce, regulated fintech, crypto, and security — payment, settlement, and security systems on AWS/GCP. Habits from that era still shape how I design tools: explicit trust boundaries, structures you can roll back when they fail, compliance logs.
Research Sites
The Ontology Lineage ko/en
2,300 years of ontology research on one static page: a lineage timeline from Aristotle to ontology-constrained LLMs, theoretical foundations, GraphRAG and knowledge-engine synthesis, an integrated reference architecture, a five-stage learning route, and a browser for all 425 papers. Collected and gap-analyzed with research-agent — the survey logged 9 gaps, a follow-up collection closed 8.
epicsagas.github.io/ontology-explorer →The Uncanny Valley of AI Writing ko/en
Why readers turn away from AI-generated text — a fact-checked dossier. All 5 cited sources verified (2 corrected), 86 papers indexed from arXiv and OpenAlex, an evidence map across 7 themes, counter-evidence in full, and 4 open gaps. The twist: readers judge provenance labels and shallow heuristics, not the text itself.
epicsagas.github.io/uncanny-writing →Featured Projects
research-agent rust
Research memory for AI agents. Indexes arXiv/OpenAlex papers into SQLite, finds coverage gaps with LLM analysis, files survey reports. Rust CLI + 18-tool MCP server.
GitHub →epic-harness rust
Multi-tool AI agent harness — 22 skills, self-evolving engine, unified memory, autonomous spec-to-PR pipeline.
GitHub →alcove rust
MCP doc server for AI agents — BM25 + vector hybrid search over private docs, tree-sitter code indexing, doc-consistency policy.
GitHub →Velith plugin
AI-native publishing system — build books like software. A 6-phase pipeline from blank page to EPUB/PDF, with 16 skills, 7 agents, 8 genre templates.
GitHub →llm-kernel rust
Rust foundation library for AI-native apps — 16-provider catalog, LLM client, MCP server, 44 local ONNX embedding models.
GitHub →Episteme rust
Software engineering knowledge graph — design patterns, refactorings, laws, and code smells. Analyze a codebase or a decision against it via HTTP API.
GitHub →More Tools
- claudy — multi-provider AI launcher
- llm-transpile — token-budget adaptive document transpiler
- obsidian-forge — Obsidian vault generator, automation, graph strengthener
- plugins — Claude Code-only plugin marketplace (skills, commands, agent configurations)
- homebrew-tap — Homebrew installs for the CLIs above
- Agent Skills — problem discovery, cognitive self-analysis, corrective routines