AI automation engineer specializing in retrieval-augmented systems, browser orchestration, and evidence-driven decision pipelines.
Most ambitious current project: OpenCodex, a systems-oriented OpenCode fork with task DAGs, semantic retrieval, TUI proof artifacts, and a stateful agent workbench runtime.
Fastest visual proof: semantic-demo, a browser-based semantic search surface over 8,406 Montgomery County business records.
My strongest work sits at the intersection of automation, retrieval, browser orchestration, data pipelines, and evidence-driven product engineering. I care about systems that can be inspected: clear runtime state, explicit failure criteria, preserved negative evidence, and tooling that helps a human operator stay oriented.
| Project | What it shows | Good first click |
|---|---|---|
| opencode-fork | OpenCodex: OpenCode fork adding task DAGs, semantic retrieval, TUI proofs, and a stateful agent workbench runtime | docs/recruiter-quickstart.md, docs/opencodex-runtime-surface.md, README.md |
| semantic-demo | Semantic search over 8,406 business records with map anchors, visual neighborhoods, and retrieval behavior you can inspect in the browser | README.md, index.html, data.dat |
| ai-operator-stack | Human-in-the-loop operating model for using AI agents, MCP tools, browser automation, memory, skills, and verification loops safely | README.md, docs/operating-model.md, docs/agent-safety-model.md |
| leadops | Lead intelligence pipeline with SQLite, vector retrieval, Streamlit UIs, mailbox parsing, data boundaries, and review queues | README.md, app.py, leadops_retrieve.py |
| permit-intelligence-dashboard | Public-record permit analytics with parcel enrichment, opportunity scoring, exports, and a static dashboard for human review | README.md, src/pipeline.py, docs/portfolio-case-study.md |
| website-audit-engine | Authorized website security and performance auditing with confidence-tiered findings, Playwright evidence capture, and batch automation | README.md, docs/authorization-model.md, audit-lead.security.test.js |
| trading-bots | Paper/experimental quantitative research infrastructure with risk controls, telemetry, proof boards, and falsification surfaces | README.md, docs/experiment-protocol.md, docs/evidence/edge_registry.md |
These are real client outcomes. mccullough.digital and mccullough.cloud are my owned delivery surfaces; the client proof is the web presence and growth infrastructure built for OnMark and ARESbuild Construction.
| Client | What I built | Proof |
|---|---|---|
| OnMark | Built the web presence, lead capture, attribution, and visibility system for a Houston solid-surface contractor, with organic/AI visibility and paid-launch metrics documented in the public case study | case study, repo proof |
| ARESbuild Construction | Built the construction website presence, project/gallery structure, trust proof, and handoff workflow behind the ARES launch-build case study | case study, delivery repo |
| Project | What it shows | Good first click |
|---|---|---|
| cloth-physics-banner | Interactive WebGL cloth banner with custom particle constraints, runtime texture generation, visual smoke checks, and a live GitHub Pages demo | live demo, README.md, src/main.js |
| soul-cycle | Browser-based procedural life-simulation demo with ES modules, J-curve mortality visualization, persistent fragments, crisis UI, and an original AI-generated soundtrack | live demo, README.md, js/main.js, assets/README.md |
| mccullough-search | Deterministic local-business website discovery with SearXNG, scoring rules, first-party verification, GitHub Actions runs, and a review queue for ambiguous results | README.md, search-lead.js, test-harness.js |
| mccullough-digital | Owned WordPress/client-delivery platform behind the client launches: custom blocks, lead capture, QA, analytics, case-study pages, and reusable delivery tooling | README.md, functions.php, templates/ |
- Evidence before claims: the best parts of these repos keep negative results visible.
- Operator clarity: dashboards, proof boards, logs, and runtime guards are treated as product surfaces.
- Pragmatic automation: I use AI where it compresses review, retrieval, drafting, or orchestration work, not as a substitute for verification.
- Safety around shared resources: browser sessions, remote edits, outreach paths, and live workflows get explicit boundaries.
Python, JavaScript, TypeScript, PowerShell, Playwright, Streamlit, SQLite, Three.js, GitHub Actions, Model Context Protocol, local embedding models, vector retrieval, browser automation, and AI-assisted development workflows.
The common thread in my work is decision systems under uncertainty:
- In trading research, that means building paper-tested risk controls and backtest infrastructure around noisy market behavior.
- In website auditing, it means separating verified user-facing issues from weak heuristics.
- In lead intelligence, it means turning fragmented public and inbox data into reviewable next actions.
- In agent tooling, it means letting multiple workers move quickly without colliding over shared tools.
- Website: mccullough.digital
- LinkedIn: linkedin.com/in/fred-mccullough
- Location: Conroe, TX