~/cebu/philippines ▸ AI AUTOMATION ENGINEER · LLM & AGENTIC SYSTEMS · 18 YRS OPS
▊
AI Automation Engineer and founder of MagnifAI — shipping production AI agents, RAG pipelines, and Model Context Protocol integrations. Built agentic systems for Wonder Group (Sydney) and leading AI & automation at Hometown Roofing & Restoration (Texas).
I ship production AI agents — into real client operations, under direct CEO oversight, then keep them running.
Founder & principal engineer of MagnifAI, an AI services agency. I architect multi-agent systems using Claude Code, OpenClaw, and a fleet of MCP connectors — work that cuts manual processing 60%+ and ramp-up time 40%.
18 years of operations and team leadership — from BPO frontline (People Support, 2006) through fraud ops at JPMorgan Chase, sales management at Options Group ($450K/mo revenue), and training at Riot Games — now applied full-time to AI engineering.
Stack: RAG · MCP · LangGraph · GoHighLevel · n8n · Claude Code · self-hosted VPS. I authored the 7-Layer AI OS framework and the Failsafe System — enforced on every engagement.
// real systems, real clients, validated end-to-end
Self-hosted AI meeting notetaker — a Chrome MV3 extension paired with a macOS native helper. Local-first, bring-your-own-key.
Vision-LLM expense system. Clients photograph receipts in a Telegram bot; structured fields land in per-client Google Sheets.
Liviti GoHighLevel Conversation AI chatbot, built under direct CEO oversight and shipped to production.
SEO content engine — a free-stack adaptation of GOLDIE Infinite Rank OS, bootstrapped autonomously via Claude Cowork.
Wired a full Zoho One MCP fleet (CRM / Projects / Cliq / Recruit / Desk) into Claude Code + Hermes across Windows and WSL.
Autonomous Discord listener that survives reboots and recovers itself — production-hardened agent plumbing.
Token-free knowledge-graph RAG engine over a 1,000+ note Obsidian second brain, exposed as an MCP server.
AI estimator converting insurance Xactimate estimates into branded customer bids — the client's top technical priority.
tmux-based reasoning engine bridging Hermes Agent to Claude Code CLI. OpenAI-compatible proxy with model switching.
Headless Chromium bot that applies to Indeed jobs in the background. Zero interruption to browsing.
Open-source coding-agent CLI for cloud and local model providers — one terminal-first workflow across every backend.
A 100x rebuild of a revert-to-mean trading EA for MetaTrader 5 — class-based, multi-symbol, panel-ready.
Dev infrastructure cutting agent token spend 60–90%. Two engines: Headroom (output compression) + RAG (retrieval).
A Zoom-style video-calling client built on Jitsi as an MV3 Chrome extension — three shipped versions.
Closed-loop AI content automation for a solar & battery company — config-driven n8n + Claude workflow.
Autonomous Hermes agent that posts a daily AI-news brief to LinkedIn at noon AEST, with a generated infographic.
Turn any tutorial or lecture video into structured study notes. Paste a URL, walk away, come back to a synthesized note.
Phone control of Claude Code on the Mac — a Telegram bot that spawns claude -p and returns the output.
// the stack runs on my hardware — every system deployed to infrastructure I administer myself
// I don't just use AI — I build the harnesses that make it ship real software
Most people prompt AI. I orchestrate agents — a fleet of specialized AI workers that decompose work, execute in parallel, verify each other's output, and ship production code without a human in the loop.
My workflow: an orchestrator decomposes the task, then specialist agents — planning, backend, database, QA, review — execute in parallel waves with verification gates (types, tests, adversarial review) between every stage. Knowledge bases persist across projects, so every build makes the next one faster.
The payoff: one engineer operating with the throughput of a team — without giving up code review, tests, or operational discipline.
My proprietary framework: Context · Orchestration · Memory · Tooling · Execution · Verification · Delivery. Enforced on every engagement.
Governance model: every agent output passes through typed gates before reaching production. Circuit breakers, rollback paths, and audit trails built into every pipeline.
Discrete task-execution agents deployed for specific operations. Each one focused, testable, and independently deployable — the smallest unit of AI leverage.