Northern CA → Portland, OR · May 2026

I build
agentic
systems
that work.

Senior Enterprise Commerce & AI Systems Lead at Insight Enterprises. I work at the seam between search, data, and AI — building the tooling that makes invisible system behavior visible. Early adopter, compulsive builder, evangelist for AI done right.

33
GitHub Projects
6mo
0 → 33 Repos
1,214
AI Conversations
3yr
Building with AI
10
Production prompts
Models evaluated

I analyzed my own AI usage.

Because if you're going to say AI changed how you work, you should be able to show your work.

3 years.
1,214 conversations.
One export file.

I exported my entire ChatGPT history and built a dashboard from it using Claude and Cowork. Monthly trends, topic clustering, model breakdown, activity heatmap. The full arc of how my relationship with AI changed — from life helper to intentional toolbox. It ends in early 2026. What came next is harder to export.

Full analysis available on request.

28×
Usage growth in 12 months
476
Active days
225
Custom GPT convos
2:1
AI/user message ratio

How I got here.

Not a straight line. Started in product data, ended up building AI infrastructure.

Dec 2022
First ChatGPT conversation
Day one of a ChatGPT Plus account that would eventually log 1,214 conversations across 38 months.
2023
Exploration phase
Recipes, ideas, questions. Treating it like a search engine with better answers. The tool was interesting but not yet a system.
Early 2024
Something clicks
Started using GPT-4 seriously at work. Enterprise search, commerce systems. The tool started feeling like a thinking partner.
Sep 2024
The inflection point
Usage jumped from ~5 conversations/month to 38, then 54. Genuine excitement about what was becoming possible.
Early 2025
Nia goes to production
Co-architected Insight's enterprise chatbot from single-prompt POC to a 10-prompt multi-model intent routing system. POC to release to ongoing ownership.
Sep–Nov 2025
0 → 20 GitHub repos
Started actually shipping. Tools, PromptKit, DiscoBot, HeirloomKeeper — one after another. Learning by building, not by reading about building.
Nov 2025
Peak: 140 convos/month
Running local LLMs, Docker infrastructure, self-built LLM router, agent-mode workflows. The personal stack is real now.
May 2026
Portland. What's next.
Moving north. Looking for the right place to bring this energy. Open to conversations.

Projects.

Real problems, real people. Some polished, some experiments. All of them taught me something.

🤖
TypeScript
DiscoBot / Disco
AI-driven e-procurement platform. Procurement shouldn't require a PhD in enterprise software. Intent-based, context-aware, built to understand what you're actually trying to buy — not just what you typed.
Agentic Procurement Enterprise
🔒 private
🛒
TypeScript
MerchGent
Multi-agent system for auditing and optimizing e-commerce merchandising. Most merchandising decisions are made with incomplete information — this surfaces what's actually happening and acts on it.
Multi-agent E-commerce AI Audit
🔒 private
💡
Python
PromptKit
A prompt engineering library built from real usage patterns — not theory. Designed for consistent results across models and contexts. Built because I kept rewriting the same things.
Prompt Engineering Python Multi-model
→ github.com/grahamton/PromptKit
🌐
Python
extraction-kit
Structured extraction pipeline for running local LLMs on real web content — no cloud APIs required. Built because the tool I needed didn't exist. The kind of project that starts as a workaround and becomes infrastructure.
Local LLM Extraction Python
→ github.com/grahamton/extraction-kit
🪨
TypeScript
Ocal
Offline-first rock identification app built for my dad — a senior beachcomber on the Pacific Coast. High-contrast, no gamification, designed for actual field use. Some apps are just for one person.
Offline-first Accessibility Vision AI
→ github.com/grahamton/ocal
📊
HTML / JS
ChatGPT Usage Dashboard
Exported and analyzed 38 months of my own AI conversation history. Built a full analytics dashboard — monthly trends, topic clustering, model distribution, heatmap. What I found changed how I think about building for other people.
Data Analysis Self-analysis Claude / Cowork
↗ Available on request
🧹
TypeScript
TishTish
ADHD clutter helper. Photograph a messy space, get LLM-generated step-by-step instructions — structured to match how ADHD brains actually process tasks, not how neurotypical productivity advice assumes they do.
Vision AI ADHD LLM
🔒 private
🏺
TypeScript
HeirloomKeeper
Log the objects that matter — family items, stories, provenance — before they're forgotten. Simple, private, built because no good tool for this existed. The right scope for a problem is sometimes very small.
Family Preservation Personal
🔒 private
📁
+24 more
And more on GitHub
OxiApp, Midimuncher, Kb-browser, exportify, compose-for-agents, agents.md, Lumen-Lab, ProcurementDisco, and experiments that didn't need names.
33 total Sep 2025 → now
→ github.com/grahamton

The actual stack.

Model-agnostic by design. I pick the right tool for each job and switch fast. What's actually running on my machine:

🔀
opencode + Copilot CLI
Model-agnostic agent layer
🐳
LibreChat + Docker
Self-built multi-model router
🧠
LMStudio
Local inference
🤖
Claude
Primary reasoning model
📱
Gemini
Mobile + multimodal
🔬
NotebookLM
Research synthesis
💻
VS Code
Primary IDE
🔍
Coveo for Commerce
Search/AI platform
🐍
Python + TypeScript
Build layer
🎹
OXI One + synths
Because MIDI is also a stack

About me.

I'm a Senior Enterprise Commerce Search & AI Systems Lead at Insight Enterprises — the person who figures out how AI and search actually work together in enterprise environments, and then gets everyone else on board.

I work at the seam between systems. Search and AI. Data architecture and product behavior. Technical reality and organizational understanding. My instinct is always to find where those diverge and close the gap — with architecture, tooling, or sometimes just a clearer conversation.

I got genuinely excited about AI before it was everywhere, then hit a wall of spinning my wheels without a real system. So I built the system. 33 GitHub repos in 6 months is what that looks like from the outside. What it feels like from the inside is finally having the right tools to think with.

Moving to Portland in May 2026. Open to conversations about what's next.

📍
Northern CA → Portland, OR · May 2026
🏢
Insight Enterprises · Sr. Enterprise Commerce Search & AI Lead
🐕
Has a dog. Responsible for at least one business idea.
🎹
Makes music with hardware synths · a different kind of system design
🪨
Built Ocal for my dad · some apps are just for one person

Let's talk.

Moving to Portland in May 2026. Looking for the right place to do serious work on agentic systems — with a team that's past the "is this real" conversation and already deep in the hard problems. Or just talking to people building interesting things.