Open to Opportunities

AI Systems Architect

Liam Petersen

Download CV liam@copyweb.io LinkedIn GitHub
0 AI Projects Built
0 Git Commits
0 n8n Workflows
$0K/mo Revenue Driven

Measurable Impact

Real numbers from production systems I built and managed across two companies.

1.2% 3.6%
3x

Conversion Rate

AI-driven content strategy I built and ran tripled site conversion rates at Kilo Health

$76K $249K
3.3x

Email Revenue

Scaled email channel from $76K to $249K in 4 months at Foxelli Group

$674K/mo
Peak

Revenue Driven

Total monthly revenue across Foxelli Group brands using AI-powered systems I built

25/wk 150/wk
6x

Video Ad Output

Scaled from 25 to 150 video ads per week with the same 4 video editors using the AI production platform I built

Selected Work

4 selected from 20+ AI projects I've architected and shipped, totalling over 10,000 commits.

588Commits
28n8n Workflows
13Airtable Tables
119Source Files

Human-in-the-Loop Production Pipeline

AI handles 90% of the heavy lifting. Humans stay in control. Planners create brands, define personas, and brainstorm ad ideas via AI. Editors execute a 6-stage pipeline (script review, hero model selection, visual direction, storyboard, video, voiceover) with human approval gates at every step.

AI-Orchestrated via n8n

28 n8n webhook endpoints handle all AI orchestration: script generation, persona extraction, storyboard creation, multi-provider video generation (VEO 3.1, Kling 3.0), voice cloning (ElevenLabs), and final Dropbox export. The Express backend proxies through an allowlisted API layer.

Learning Flywheel

Editor feedback captured after every job. Claude analyzes patterns and compiles structured learning briefs that are auto-injected into future script generation. The system improves with every ad produced.

Production Safeguards

Clerk auth with role-based routing. Endpoint and table allowlisting so only pre-approved n8n endpoints and Airtable tables are accessible. Tiered rate limiting (60/20/5 req/min). Adaptive polling with stuck job detection and automatic reset.

React 18 TypeScript Express 5 n8n Cloud Clerk Auth Airtable VEO 3.1 Kling 3.0 ElevenLabs Gemini 2.0 Claude API Tailwind CSS shadcn/ui TanStack Query Framer Motion
Production tool deployed on Replit, used daily by Foxelli ads team
3Core Workflows
6Python Tools
<$7Per 30-Day Campaign
4Social Platforms

ART Framework Architecture

.agent/ folder with AGENT.md brain (404 lines), modular skills (Blotato scheduling, Modal deployment), and 3 declarative workflows: 30-Day Campaign, On-Demand Generation, and YouTube-to-LinkedIn automation.

Multi-Provider Abstraction

Provider registry routes to Google AI Studio (free tier), Kie AI, or WaveSpeed based on model and availability. Image generation via Nano Banana Pro, video via Veo 3.1 / Kling 3.0 / Sora 2 Pro. Automatic fallback when a provider goes down.

Full Campaign Pipeline

Brand discovery interview generates BRAND.md voice file. Creates 30 Airtable records with prompts and captions. Batch generates all images. Schedules to TikTok, Instagram, YouTube, and LinkedIn via Blotato. Human reviews at every stage and nothing auto-publishes without approval.

Serverless Automation

YouTube-to-LinkedIn pipeline runs daily on Modal.com (~$0.30/month). Monitors RSS feeds, extracts transcripts, generates branded infographics, writes captions in creator's voice, and posts automatically. Uses persistent volumes for state tracking.

Python Claude Code Google AI Studio Kie AI WaveSpeed AI Gemini 2.0 Flash Airtable Blotato Modal.com
99Ad Principles
28n8n Nodes
4,450Lines of Code
$0.40Per Full Cycle

Strategy-First Generation

Every ad is grounded in a named advertising principle from a library of 99 (@The_AdProfessor) and cross-validated against David Ogilvy's 24 lessons + 38 rules. Claude Sonnet generates 3 concepts per batch, each with a different principle, clever idea, headline, image prompt, and complete Facebook ad copy.

Competitor Reverse Engineering

Upload a winning competitor ad and the system explains WHY it works (Ogilvy analysis across 10 dimensions), extracts its exact spatial layout as a JSON spec (headline position, product placement, stat badge zones), then rebuilds it for your brand while preserving the winning formula and swapping all content.

Intelligent Ad Matching

Curated library of winning ads tagged across 10 dimensions (~200 controlled vocabulary tags). Weighted Jaccard similarity scoring (campaign_goal 3x, industry 2.5x, tone 2x) matches brands to the best-performing templates. All running client-side with zero API calls.

Layout Zone System

Solved the problem of image AIs ignoring spatial instructions. Each layout zone's position and size gets converted to natural language ("across the top of the image, centered, a large prominent headline..."). Zones sorted top-to-bottom for natural reading order. Client-side canvas logo compositor for pixel-perfect brand placement.

React 18 n8n Cloud Claude Sonnet Nano Banana Pro Kie.ai Kling 3.0 Sora 2 Pro Airtable Python Ogilvy Framework
5Specialized Agents
4Daily Cron Triggers
3-LayerMemory System
TelegramCommand Center

Scout: Recon Specialist

Daily B2B prospecting with weighted ICP scoring rubric (5 dimensions: Team Size 25%, AI Readiness 25%, Budget 20%, DM Access 15%, Pain Signals 15%). Auto-enriches via Apollo.io + Hunter.io. Tracks pipeline across 11 stages with 14-day stale lead auto-flagging.

Wordsmith: Content Engine

Weekly LinkedIn/X content batches using Hook-Retain-Reward structure. Enforces 3:1 value ratio. Rotates content types (How-To, Myth Bust, Case Study). Monitors engagement signals and flags ICP decision-maker interactions.

Closer: Sales Operations

Complete sales workflow from A-C-A outreach drafting through 3-touch follow-up sequences to proposal generation. Pre-call SPIN question briefs. Post-call proposals using client's own language from interviews. 8-step onboarding automation.

Architect: Client Delivery

6-phase delivery engine: pre-diagnostic research, interview prep, transcription processing, scorecard generation, blueprint presentation, and evolution monitoring. Generates 90% complete first drafts of all client deliverables.

Command Center: Orchestrator

5:30 AM data sync, 7 AM morning brief, 12 PM midday pulse, 8 PM evening wrap, all via Telegram. Cross-agent coordination (Scout-to-Closer handoff, Wordsmith-to-Scout engagement signals). Voice note routing with transcription. Auto-escalates overdue tasks.

Claude Code SKILL.md Architecture Cron Scheduling Telegram Apollo.io Hunter.io 3-Layer Memory Cross-Agent Protocols

Tech Stack

Tools and technologies I work with daily.

AI Models

Claude Opus / Sonnet GPT-4o Gemini 2.0 Flash ElevenLabs Groq Whisper

AI Video & Image

VEO 3.1 Kling 3.0 Sora 2 Pro Nano Banana Pro Nano Banana 2 Kie AI

Automation

n8n Cloud Claude Code Agent SDK Cron Scheduling Modal.com

Frontend

React 18 TypeScript Tailwind CSS shadcn/ui Framer Motion TanStack Query

Backend & Data

Express 5 Node.js Python Airtable SQLite

Infrastructure

Replit Clerk Auth Telegram API Dropbox API Apollo.io

The Journey

From copywriter to AI architect in under 4 years.

2022

Junior Copywriter

Started in marketing. Writing copy, running email campaigns, learning what makes customers convert.

2023-2024

Head of AI Copy

Built AI into the content workflow. Tripled conversion rates, scaled email revenue 3.3x. Started automating everything.

2025-Present

AI Strategy Manager

Architecting full-stack AI systems. Production platforms, autonomous agent teams, and AI operating systems for marketing teams.

About

Liam Petersen

I believe the best AI systems keep humans in the driver's seat. Every system I build automates 90% of the repetitive work while keeping the core human skill in charge. In marketing, that matters more than anywhere else because brand voice, creative instinct, and customer empathy can't be fully automated.

I went from Junior Copywriter to AI Strategy Manager across two companies in under 4 years. I'm not a traditional developer. I'm a marketer who taught himself to architect and ship AI-powered software using Claude Code, n8n, and AI APIs.

I've driven real results: tripled conversion rates (1.2% to 3.6%), scaled email revenue from $76K to $249K in 4 months, and built a self-improving AI content pipeline that gets smarter with every ad it produces.

Certifications

Claude Code in Action (Anthropic) Building with the Claude API (Anthropic) Introduction to MCP (Anthropic) MCP: Advanced Topics (Anthropic, In Progress)