Marketing OS
One system runs three client engagements.
Multi-client marketing operating system with shared skills, isolated context, and enforced quality standards.
One repo runs the entire consulting practice.
I run marketing engagements for three clients: a behavioral health practice, a mid-market HCM consulting firm, and a regional trucking company. Each one needs voice profiling, brand positioning, content production, and SEO work. The deliverables overlap, but the context is completely different.
Cortex is the system I built to run all of it. Reusable skills handle the logic (write a blog post, run an SEO audit, build a sales deck). Per-client folders hold the context (voice, brand, sources). Shared standards enforce quality across every deliverable the system produces.
I built the operating system for the work.
A three-layer system with hard boundaries.
Skills never contain client data. Client folders never contain logic. Standards sit above both and get read by everything. Meaningful changes are version-controlled in git.
Shared standards
Writing Standards
80+ anti-pattern rules enforced on every piece of prose.
Voice Templates
16-section profile format applied to every client.
Output Schemas
Skill-map spec, platform overlays, context format.
Stateless skills
Content Engine
Radar, orchestrator, and format-specific generators.
SEO Audit
Six modules plus a synthesis roadmap builder.
Client Pipeline
Intake interviews, voice profiling, brand development.
Sales Deck System
Positioning to slide deck, end to end.
Behavioral health practice
HCM consulting firm
Regional trucking company
Consistency is the actual hard problem.
Context lives in your head
Every client has a voice, a positioning, a set of sources. None of it persists between sessions. You re-explain the client to the tool every time you sit down.
AI output drifts without constraints
Prompt a model for a blog post and you get a blog post. Prompt it again tomorrow and you get a different one. No enforced standards means every deliverable is a coin flip.
Agency workflows break at one
Traditional agency process assumes multiple specialists with shared tooling. A solo operator running three clients needs the same consistency without the headcount.
Cortex solves all three by making context persistent, logic reusable, and quality standards automatic.
The system in production.
These are the current numbers from the repo.
3
Active client engagements
~30
Skill and module files
6
SEO audit modules + synthesis
80+
Documented AI anti-patterns
16
Voice profile sections
8
Shared standards and templates
From intake call to enterprise sales deck.
A mid-market HCM consulting firm needed positioning, a messaging guide, a voice profile, and a sales deck for a high-stakes presentation to partner-channel sales reps. The entire pipeline ran through Cortex.
Conversational intake
A structured interview captured the firm’s positioning, audience, competitive landscape, and presentation goals. The intake skill converted the transcript into a client profile that every downstream skill could read.
Positioning framework
From the profile, the system produced a messaging guide: value propositions ranked by audience segment, proof points mapped to each claim, objection-handling language for the sales team.
Voice profile
A 16-section voice profile captured how the firm talks (and how it doesn’t). Vocabulary, sentence rhythm, formality level, banned patterns. Every content skill reads this before generating a word.
Sales deck production
The deck skill assembled positioning, proof points, and voice into a presentation built for a room of partner-channel sales reps. Design system, master template, and speaker notes included. The deck was used in the actual meeting.
“Hire me and I bring the system with me. Voice profiling, content production, SEO audits, and quality enforcement that runs from day one.”
I designed how the tool should work, then built it.
The writing standards file has 80+ anti-pattern rules. Every prose-generating skill declares that it requires those standards and reads them before generating a word. A voice review gate blocks content production when more than a third of a client's voice profile is unresolved.
The system enforces taste. Anyone can get a model to produce text. Getting it to produce text that holds a specific client's voice, clears a quality bar, and does it the same way on Tuesday as it did on Friday requires architecture.
Three live clients run through it every week.
A senior PMM, growth, or head of marketing seat where the function is mine to own.
Available immediately. Remote-first, open to Chicago onsite. Targeting healthtech, B2B SaaS, and cybersecurity at Series B through public.