Ryan Fortin — AI Marketing Technologist

Experience

2015 — Present Raleigh-Durham, NC

Lenovo / Global SEO Program Manager

AI Agents, Marketing Automation & GEO 2024 — 2026

Designing and building Lenovo's first OpenClaw AI agent for the worldwide eCommerce marketing team: a platform built from scratch that automates repetitive marketing workflows, generates and optimizes content, and surfaces data and reporting insights.
Partnering directly with the full WW marketing organization to customize the agent to each sub-team's needs, leading change management and training to shift teams from manual execution to managing AI-assisted workflows.
Developed a comprehensive prompt-engineering library for AI-generated content, establishing best practices for Lenovo's global content teams.
Partnered with Lenovo IT to develop an SEO Agent that increased operational efficiency by 90% through automated technical audits and recommendations.
Pioneered Generative Engine Optimization (GEO) methods to optimize Lenovo's visibility in ChatGPT, Perplexity, and other AI search platforms.
Created AI content-generation workflows integrating multiple LLMs for scalable, SEO-optimized content production.
Established monitoring systems to track Lenovo brand mentions and product recommendations in AI-generated responses.

Global SEO Program Leadership 2015 — Present

Manage multi-million-dollar SEO-driven eCommerce revenue across 30+ countries with consistent YoY growth.
Hired and lead a world-class international SEO team supporting stakeholders across the globe.
Drive strategic investment decisions for resource allocation and ROI optimization across global markets.
Established a data-driven SEO culture through cross-functional evangelization and training programs.

2012 — 2015 San Diego, CA

Covario / iProspect / Sr. SEO Strategist → Global SEO Lead

Rose from Sr. SEO Strategist to Global SEO Lead through Covario's acquisition by iProspect.
Led SEO for Lenovo as the flagship client, an international program spanning 14+ countries and the engagement that led to joining Lenovo in-house in 2015.
OMMA Best SEO Agency of the Year (2013) · eTail Search Marketing Award (2014).

2010 — 2012 Manhattan, NYC

Catalyst Online / GroupM Search / Search Manager, SEO

Delivered SEO campaigns for Fortune 500 clients (Audi, Pfizer, GSK, Novartis) inside GroupM, WPP's media investment arm.
Owned end-to-end execution of client organic search campaigns, building winning strategies from client business objectives and the latest search-algorithm knowledge.

2008 — 2010 Rochester, NY

Archer Communications / Director of Web Technology

Led technology across all of the agency's web projects: site development, infrastructure, database architecture, and project planning.
Owned SEO and site analytics for every web build. That was the start of 17+ years in search, with engineering first and marketing layered on top.

AI / ML Projects

Fine-tuned LLM for Generative Engine Optimization

Qwen GEO Fine-Tune

Qwen3-30B fine-tuned into a GEO specialist: 5,000+ custom training examples baking Conversation Optimization, my own GEO strategy, directly into the model weights. Trained with 4-bit quantization, gradient checkpointing, and aggressive LoRA parameters; deployed with automated checkpoint management and GGUF conversion.

Python · Unsloth · PyTorch · LoRA/QLoRA · Runpod A40

huggingface.co ↗

Automated LLM optimization

Evolutionary Model-Merging System (M2N2 Breeder)

An automated breeding system that evolved a champion Qwen3-8B over 781 generations and 749 merges across a 24-model population of domain fine-tunes. Tournament selection, niche-based fitness, zero additional training. Beat base Qwen3-8B on 6 of 8 benchmarks (+3.4% avg across MMLU, GSM8K, ARC, HellaSwag, WinoGrande, MathQA), validated with EleutherAI's lm-evaluation-harness.

Python · mergekit (SLERP, DELLA) · lm-evaluation-harness · PyTorch

huggingface.co ↗

AI safety & interpretability research

Activation Steering (4ort-kardashev)

Interpretability research that installs behavior inside the model itself, at the activation level, using sparse-autoencoder feature discovery and an original multi-layer clamping technique. The steered behavior survives a hostile system prompt explicitly designed to override it. For enterprises, that's the difference between a policy the model is told and a policy the model has: brand-safety guardrails that can't be prompt-injected away.

PyTorch · SAEs on Qwen3.5-27B · frozen-metric eval suite · single RTX 4090

research deep dive →

Capabilities

AI capability, proven in production. Every service below points at something already built and running. Pairs with the SEO practice: see Conversation Optimization under SEO Expertise.

Capability / 01

AI Agents for Marketing & Operations

Design, build, and deploy production AI agents that take over repetitive execution (content operations, reporting, and workflow automation), customized to how your team actually works.

Mission Control →Lenovo's OpenClaw agent →

Capability / 02

Agent-Ready Infrastructure

Make your product, data, or content operable by AI agents: MCP servers, agent-first APIs, and structured-data layers that turn "we have an API" into "agents can work here."

21 properties of proof →@4ort/cli on npm ↗

Capability / 03

Custom Models & Fine-Tuning

Domain-specialist LLMs trained on your data and your methodology, covering dataset design, fine-tuning, model merging, and benchmarked evaluation through to deployment.

GEO fine-tune ↗M2N2 breeder ↗

Capability / 04

SEO Automation Systems

Audit agents and optimization pipelines that replace manual SEO execution: continuous technical auditing, automated on-page fixes, and enterprise-scale workflows.

The 90% SEO Agent →LayerSEO & agenticSEO →

Capability / 05

Autonomous Content Pipelines

Content factories with quality gates. Video, audio, and long-form text are generated, curated, and published without a human in the loop.

~214 videos →100+ tracks →2 books →

The 4ort Galaxy

A solo-built, self-hosted network of 20+ interconnected AI properties federated under one identity, where autonomous agents are the primary users. One knowledge graph grounds every agent in verified facts; one hub orchestrates them all.

4ort.ai

Mission Control

Multi-tenant AI chat platform and sole identity provider that orchestrates every property's MCP toolset, giving a single agent authenticated hands across the entire stack. Ships its own toolkits: LayerSEO, the 32-tool agenticSEO MCP, a custom self-hosted code interpreter, and a secret firewall.

Node/TypeScript · React · MongoDB · OIDC · MCP

4ort.xyz

Knowledge Graph & Worldwide Trend Engine

A public, Wikidata-grounded graph of ~40M entities that grounds every agent in verified facts and measures worldwide trending topics in real time, continuously scanning the full GDELT global-news firehose to mint brand-new entities within 15 minutes of them appearing in world news.

Python/FastAPI · Postgres (~313GB) · Redis · SSE

4ort.news

Worldwide News Brain

Monitors global news across 65+ languages every 15 minutes, surfacing emerging stories and newly-minted entities. A per-user OAuth MCP server lets an agent read, summarize, and manage the feed autonomously. Real-time brand, competitor, and market monitoring.

Django · Postgres/Elasticsearch · GDELT · MCP + OAuth

4ort-docs / sheets / slides

Agent-First Content Suite

Documents, spreadsheets, and decks built agent-first (REST API + MCP toolset + human editor), with AI per-slide editing that round-trips "fix slide 3" in ~3 seconds. Agents produce and revise marketing collateral programmatically.

TypeScript · Express · SQLite · MCP

4ort.shop

World Product Index

An API-first crawler that mapped 191,000+ ecommerce domains and 5,300+ live product-catalog endpoints into a structured JSON-LD entity layer. The world's commerce, indexed for agents.

Python · async httpx

4ort-deep-research

Agentic Deep Research

A deep-research agent that swaps every vendor tool for self-hosted galaxy infrastructure, then benchmarks it: 84% pass@1 on a 50-task FRAMES subset (up from 82%), at or above vendor-tooling baselines. It runs as a 5-worker concurrent pool across four depth tiers, from ~10-second answers to multi-hour research.

PyTorch overlay · Redis · systemd

4ort-cli

The Galaxy CLI

One command-line client for the whole galaxy, published on npm as @4ort/cli: site hosting, mail, knowledge-graph search and Q&A, smart web search, video publishing, and 1Password-style secrets injection (4ort run feeds vault secrets to child processes only; plaintext never touches disk). An mcp mode turns the CLI itself into an MCP server for any agent.

TypeScript · commander.js · v0.7.0 on npm

npmjs.com ↗

// the rest of the constellation

4ort.stream

Autonomous Video Pipeline

~214 videos published with no human in the loop, each scripted from cited knowledge-graph facts, narrated, moderation-gated, and auto-published overnight, every night.

4ort-kardashev

Activation-Steering Research

Interpretability research steering an open LLM at the activation level through SAE feature discovery and multi-layer clamping. The steered disposition survives a hostile system prompt that explicitly bans the topic.

4ort.nexus

Search for Agents

A search engine built for AI agents, not eyeballs: a 32M+ page Common Crawl corpus cross-referenced into the entity graph, served as a typed API with agent-friendliness verdicts per source.

4ort.io

The Inference Router

An OpenAI-compatible multi-provider LLM gateway, plus the Council: fan a question out to a panel of frontier models and synthesize their collective answer behind one model id.

4ort.live

The AI Radio

A 24/7 AI radio station where LLMs invent the prompts, a music model synthesizes original tracks around the clock, and everyone tunes into the same synced stream. 100+ tracks and counting.

4ort-music

The Music Factory

The generation engine behind the radio. LLMs invent each song's theme, BPM, key, and lyrics; ACE-Step synthesizes it on CPU 24/7; an ~80% cull rate means only the keepers reach the 100+ track library.

4ort-books

The Novel Engine

An AI novel-generation pipeline with two books produced (including an 8-chapter novel with a full audiobook), grounded in dual knowledge graphs with calibrated prose-quality gates.

4ort-tv

Long-Form AI Video

Topic → finished long-form MP4, end to end: LLM script, AI slide generation, neural narration, original AI music, GPU composition.

4ort.mov

Video Platform + Render Engine

The galaxy's self-hosted video platform plus a video-factory render engine, fed by the autonomous pipelines through a single CLI publish path.

4ort.life

Personal KG + Life Simulator

Your private knowledge graph with a life simulator on top: a nightly "Tomorrow" simulation briefs you on your own next day. The private twin of the public graph.

4ort.win

Public-Opinion Simulator

Feed it any document; hundreds of AI agent personas simulate public reaction hour-by-hour, and it can substitute your actual network into the simulation.

4ort.info

The Foresight Orchestrator

A prediction & ideation engine whose bisociative "spark" ideation collides distant domains from the knowledge graph into foresight, campaign, and new-product reports.

4ort.net

The Personal Namespace

Identity infrastructure for humans and their agents: you@4ort.net email on a self-hosted mail server, you.4ort.net static hosting pushed via CLI, encrypted credential provisioning.

4ort.dev

The Launchpad

The galaxy's self-hosted code backbone. Every property repo lives on a private git instance, gated by secret-scanning pipelines. Every repo is a rocket.

20+ properties, one builder ~40M entities in the public graph ~214 videos published autonomously 84% pass@1 on FRAMES, all self-hosted 191k+ ecommerce domains mapped 100+ original AI music tracks 2 AI-written books

Technical Expertise

SEO — 17+ Years

Enterprise SEO · E-commerce SEO · International SEO · Technical SEO · Generative Engine Optimization · Data Modeling & Forecasting · SEO Program & Team Leadership · Structured Data (Schema.org / JSON-LD) · SEO Automation & Agents

AI / ML

Large Language Models · Transformer Architectures · Fine-tuning (LoRA/QLoRA, SFT) · Prompt Engineering · Agent Design (MCP, Tool Calling) · RAG Systems · Vector Databases · Model Training (Unsloth, PyTorch) · Multi-modal Models

Generative Engine Optimization

ChatGPT Optimization · Perplexity AI · Claude Search · AI Answer Engines · Citation Optimization · Structured Data for AI · AI Search Ranking Factors

AI Platforms & Models

Anthropic Claude (Opus, Fable 5) · OpenAI ChatGPT 5.5 · Google Gemini · DeepSeek V4 Pro · GLM-5.1 · Kimi K2.6 · Qwen · AWS Bedrock · Model Context Protocol (MCP) · Ollama · LM Studio · Claude Code

Marketing Ops & Automation

AI Agent Design for Marketing Workflows · Marketing Automation · Campaign & Content Operations · Change Management & Team Enablement · Cross-Functional Stakeholder Management · Workflow Automation · API & Data-Flow Integration · Salesforce Data

Data & BI

Tableau · SQL · Data Visualization & Dashboarding · Large-Dataset Analysis · Insight-to-Action Reporting · Marketing Performance Analytics

Marketing & SEO Tools

Adobe Analytics · SEMrush · BrightEdge · Scrunch AI · DataForSEO

Education

Rochester Institute of Technology · B.S. New Media Information Technology, 2003–2007

Recognition

Industry speaker at Digital Marketing Phoenix, Digital Marketing Dallas, and Mobile Commerce Summit. Published on SEO and AI-search optimization strategies. OMMA Best SEO Agency (2013) · eTail Search Marketing Award (2014).

● OPEN CHANNEL

Let's talk.

linkedin.com/in/ryanfortin ↗ huggingface.co/ryanfortin ↗