Evidence Brief
AI Product Capabilities & Infrastructure Work
This document summarizes work I've completed over the past 9 months—both at Cedar Gate and independently—that demonstrates readiness for an AI-focused role within the product team. Every claim cites a verifiable source. Every number has been confirmed against the original system or document.
1,573
Knowledge Graph Notes
80+
AI Skill Specifications
104K
Words Written (2 Books)
4,661
Semantic Tags Indexed
What I've Already Built for Cedar Gate
The AI work didn't start as a side project. It started as the natural next step of the same systems thinking I've been applying to marketing infrastructure for five years.
AI Center of Excellence Business Plan
75-file production-ready specification with SQLite state database
Complete platform architecture for governed AI at Cedar Gate. Includes financial model projecting $93K Year 1 savings ramping to $141K–$167K/year steady state. Three operational Packs defined (Bundles BI, CMI Readouts, Relationship Intelligence) targeting $306K/year in manual work. Technology architecture specifies a Context Orchestrator kernel with versioned contracts, Brain Adapter abstraction for vendor-agnostic LLM switching, Evidence Bundle generation for audit trails, and a Tool Firewall enforcing ALLOW/ASK/DENY policies. Phased 24-month roadmap included.
Template Manager Reimagining
Co-created AI product vision with the product team (February 2026)
Contributed to three progressive concepts replacing the manual Excel round-trip workflow with AI-driven reporting: Conversational Report Builder (~5 min to report), Pin/Compose/Publish (GenBI discoveries become building blocks), and Living Report Canvas (persistent, interactive, collaborative). 8 architecture diagrams, 17 API endpoints, 6 database models, video walkthrough, and prototype spec delivered. Named as TM expertise stakeholder in the concept document.
Marketing Infrastructure (Existing Role)
GTM Reporting Infrastructure
Built integrated reporting connecting marketing activities to pipeline and revenue impact—attribution models and KPI dashboards for executive decision-making.
Brand Consolidation
Led unification of Cedar Gate's digital brands following multiple acquisitions into a cohesive digital presence. Systems integration at organizational scale.
"The AI COE plan wasn't a side project. It was the natural output of the same systems thinking I've been applying to marketing for five years. Analytics infrastructure → reporting pipelines → attribution models → AI-driven automation. Same trajectory, not a career change."
What I've Built Independently
Starting from zero programming experience in May 2025, the following systems were designed, built, deployed, and maintained over 9 months. Each compounds on the infrastructure of the ones before it.
Graph RAG Knowledge System — 1,573 notes, hybrid search, MCP server
Custom pipeline processing AI conversations into a searchable, interconnected knowledge graph. 4,661 tags emerged organically (power law distribution). 768-dimensional embeddings via local bge-m3 model. Hybrid retrieval (FTS + semantic + graph) with Reciprocal Rank Fusion. Accessible via MCP Server from any compatible client. 46% note growth and 97% link growth over 6 months.
negrini.io — Knowledge publishing platform (Next.js 16, 48 build phases)
Custom-built website: 122 artifacts across 21 thematic threads, 8 book chapters, 8 guided learning trails, 50+ glossary terms. Velite content pipeline (MDX → JSON), Fuse.js search, Mermaid diagrams. Lighthouse performance: 94–99.
Claude Code Skills Ecosystem
80+ prompt engineering specifications across video production, brand systems, business intelligence, knowledge management, development methodology. Each skill composes into complex workflows.
ACE Framework
Comprehensive reference specification for context management in multi-agent systems. Synthesized from 24 systematic research reports covering transformer mechanics through production deployment.
Two Books (~104,000 words)
"Build the Thing that Builds the Things" — technical memoir weaving AI architecture with ADHD-cognition discovery. 8 chapters (Book 1) + 52 constellation files (Book 2).
Local LLM Hybrid Router
95–99% cost reduction ($75–150/mo → $0.12–0.25/mo). 80% of queries handled locally, 20% routed to Claude. Complexity-based routing. ROI: first month.
The Compound Effect
| Project | Duration | Why Faster |
|---|
| Project 1 (Graph RAG) | 12 weeks | Discovery, learning curve, building infrastructure |
| Project 2 | 6 weeks | Templates reduce discovery time |
| Project 4 | 2 weeks | Patterns proven, decisions documented |
| Project 10 | 2 days | Infrastructure mature, automation handles most work |
Why This Combination Matters for Product
Cross-Domain Synthesis
Healthcare domain knowledge + AI architecture + go-to-market thinking. Not switching lanes—converging them. The COE plan translates between technical architecture and business stakeholders because I've spent a decade doing exactly that in marketing.
Infrastructure-First Thinking
Every system I build creates foundations that compound. The Graph RAG enables the website enables the books enables the skills. The COE kernel enables Pack 1 which accelerates Pack 2 which makes Pack 3 trivial. This isn't feature-building—it's platform thinking.
Context Engineering from First Principles
Not just using AI APIs—understanding why context management, persistent memory, and orchestration patterns matter architecturally. The 24-report research corpus and ACE Framework demonstrate depth beyond tool usage.
Technical ↔ Business Translation
The COE plan isn't a technical spec that needs business translation. It's both simultaneously—financial models alongside architecture decisions, governance frameworks alongside API specifications. Proven ability to hold both lenses.
Systems Thinking Roots
Sociology degree = understanding how people, institutions, and communication streams interact. This is foundational for designing AI products that humans actually use. Technology that doesn't account for how people work in organizations fails regardless of its technical merit.
Velocity
9 months. Zero to: 1,573-note knowledge graph, 122-artifact website, 75-file business plan, 104K words across two books, 80+ skill specifications, and the ACE reference framework. This is what infrastructure-first compounding looks like when applied to learning.
Credentials
VP, Digital Marketing — Cedar Gate Technologies
May 2021 – Present
VP Digital Marketing — CommCreative
Nov 2017 – Jan 2021
Digital Marketing Strategist — Wakefly
Jan 2015 – Nov 2017
MBA, Marketing — Clark University, Worcester MA
2011
BA, Sociology — Massachusetts College of Liberal Arts, North Adams MA
2009
Explore the full body of work:
negrini.io
A Note on Where This Came From
I discovered that the infrastructure patterns I build for AI systems are the same patterns I've been building for my own cognitive architecture my entire life. Context windows are working memory. RAG is external memory. Agent orchestration is bounded context coordination. That's not a metaphor—it's structural equivalence. The reason I understand what AI infrastructure needs is that my ADHD brain has needed the same infrastructure all along. I built for myself first, and it turned out to be what AI systems need too.
Full exploration: negrini.io/book