Infrastructure I've Built

Most of what’s on this page is a single system with a name and a purpose. This one is different — it’s the actual infrastructure underneath everything else here: the real automation layer that runs, schedules, and maintains the rest of this portfolio and the operational work behind it.

Why This Exists

Every project on this site produces something — a game, a compliance tool, a data pipeline. This is the layer that makes those things keep running without me manually operating each one: publishing schedules, service orchestration, cross-session memory, and the tooling that lets an AI agent actually act on a real codebase instead of just describing what to do.

This isn’t one project. It’s a real, currently-running constellation of MCP (Model Context Protocol) servers — the same protocol that lets an AI assistant call real tools against real systems, not just generate text about them.

Named Systems

DuggerBot — a coding-agent orchestration layer. Dispatches bounded coding tasks to a local agent, tracks directive and project state across sessions, verifies real test floors rather than trusting agent self-reports, and runs local inference for fast lookups that don’t need a full model call. 25 tools, 258/0/0 certified floor. Evolved from an earlier personal-assistant project of mine (PrivyBot) into this current, more specialized form.

RFD_YT_Engine — YouTube content scheduling and catalog management. Two-store design: a full local mirror of every video on the channel for fast queries, and an append-only stats cache for trend tracking over time. Detects scheduling collisions, manages batch rescheduling with a preview-before-apply safety pattern, and enforces its own architectural boundaries with structural tests — two of its internal modules are physically forbidden from importing each other, and that rule is tested, not just documented. 24 tools, 199/0/0.

rfd-blog-engine — WordPress and Dev.to publishing, with atomic scheduling (a reschedule either updates both the CMS and the internal calendar together, or neither, never a partial write) and an idempotent daily sync job that’s safe to re-run without creating duplicates. 184/0/0, with an honest, currently-open finding baked into its own state tracking: the blog has zero tags and zero featured images across every post — a real gap, already measured, not yet closed.

RFDHugoMCP — the tooling this exact site is built and deployed through. Read, write, and build operations against the live Hugo site, with git-based deploy triggering the actual GitHub Actions pipeline. This page you’re reading was written using it.

rfd-gateway — a service registry and routing layer sitting in front of the other tools, with a three-tier fallback (live service → cache → static config) so a single service restart doesn’t take down everything that depends on it. 20-tool catalog.

rfd-memory — cross-session persistence. Lets an AI assistant retain project context, architectural decisions, and state across conversations that would otherwise start from zero every time.

RFDStudioMCP — live inspection tooling for the game engine behind RFDGameStudio — load a game, inspect live state, call functions directly, run the test suite, simulate balance scenarios. 11 tools.

Generalized Systems

Some of this infrastructure was built for specific operational environments where the details aren’t mine to publish, even generalized — but the capability is real, and worth being honest about rather than omitting entirely.

A real-time dashboard automation layer for a contact center dialer platform — reorganizes a live agent monitor into team-grouped views using CSS-first, non-destructive rendering that survives the underlying platform’s own high-frequency updates.

A list-performance scoring system for outbound dialer campaigns — evaluates and ranks call lists by performance to help prioritize which lists are worth continued use.

A personal finance data pipeline — parses bank statement PDFs (no export API available) into a structured, queryable transaction database. 157 real statements, nearly 11,000 parsed transactions, 9 tools, 115/0/0. Not featured here in detail — it’s built on personal financial data — but real, working, and the same engineering discipline as everything else on this page.

The Actual Number

Across the systems above: roughly a dozen real MCP servers, several hundred individual tools, most enforced by their own certified test floors rather than assumed to work. This is the same discipline applied everywhere else on this site — real numbers, real state files, real verification — turned toward the infrastructure that keeps the rest of it running.

This page describes real, currently operating systems. Numbers are dated to when they were last directly verified, not assumed current indefinitely — infrastructure like this changes fast.