Portfolio

Platform architecture, documentation strategy, and end-to-end ownership of docs.domaintools.com for a 150-person B2B security intelligence company serving 800+ customers. OpenAPI-as-contract across four API product lines, CI/CD with GitOps-managed deployment, SME contribution workflows across 4 engineering teams, and AI-integrated quality control via DDT. Unified 120+ documents from scattered sources into a single MkDocs-Material platform with Swagger UI, search, and release-cycle integration. Produces auto-generated structured context files (llms.txt, LLM context files) covering all API products and data schemas via CI/CD for use in LLM tooling and retrieval workflows.

Production MCP server for deterministic quality control in LLM-assisted documentation. Integrates Vale, markdownlint, spaCy, and custom linters as callable MCP tools, giving LLM writing assistants a structured, auditable quality gate, the mechanical enforcement layer beneath the DomainTools documentation platform.

A git-native knowledge system that lets AI documentation agents improve safely: every change to what the agents know is human-reviewed via pull request, so nothing degrades silently. Four-tier memory architecture with an explicit consolidation loop that captures lessons from completed work and promotes them into reusable skills and standing rules. Deployed at DomainTools across multiple agent tools, sitting beneath DDT and the documentation platform as the workflow-memory layer.

Linter for LLM agent control labels: route names, tool names, enum values, environments, and approval targets. Detects confusable label pairs weighted by what the label controls, the same reliability concern that drives DDT's quality gate, investigated independently through telegraph codebook history. Research published alongside a companion essay on the Wireword blog post.

OpenAPI specification for real-time and batch feed delivery, including examples, reusable schemas, error contracts, and download workflows.

API reference documentation for DomainTools data products and endpoints, covering the public product surface across four API product lines.

Developer guide and API reference: installation, authentication, usage patterns, and method reference for Python SDK users.

Published essay. Systematic review of telegraph techniques (compression, error correction, codebook design) applied to LLM agent architectures, finding that most practices survived as information-management fundamentals, and the one actionable gap is redundancy in control-label design. Cites recent arxiv research on LLM tool-selection sensitivity and adversarial tool-name attacks. Paired with the Wireword linter as a proof artifact.

Published essay. Argues that the power of an abstraction layer comes from committing to it entirely: partial adoption breaks the model. Traces this principle from Strachey's 1965 macrogenerator through Dijkstra, Kay, and into current LLM-assisted work. Closes with first-person investment in the discipline of learning LLM failure modes.

Practical workflow essay. Documents the Keep a Changelog [Unreleased] section pattern as structured LLM session state, a small, version-controlled working-memory document that reduces cold-start orientation cost and preserves accruable context across sessions.

Local CLI and LCARS TUI for tracking LLM token and cost data across Claude Code and Cline sessions. Reads session files, deduplicates against a local ledger, reports cost and token breakdowns across multiple providers (Anthropic, OpenAI, Bedrock, Google, Meta, Mistral).

Single-instance AWS EC2 personal cloud lab: Tailscale-only access, CLI lifecycle tooling, persistent state volume, snapshots, restore drills, runbooks, and cost guardrails. Designed to be operated from lightweight ARM/Linux nodes.

This site: Hugo static publishing with GitHub Actions deployment. A vintage computing pipeline drives emulated VAX and PDP-11 systems via SIMH/pexpect to generate a man-page resume artifact, a technical signal of systems-level documentation thinking.

Peer-reviewed paper. Co-authored historical analysis tracing endpoint security management from the 1970s through the COVID-19 work-from-home transition, complementing a 1,000-hour SOC ethnography.

Comprehensive history of internet unique identifier creation and administration from ARPANET origins through 2017. Jointly commissioned by ICANN and Google for policy deliberations on IP governance and ownership.

Peer-reviewed article on the co-development of cryptography and networking, tracing the relationship between edge devices and the evolution of network security.