Sales & Marketing Portfolio · V1

What I build and run.

I'm James. I build with AI. Finance degree, then an operations leadership role running a family Mercedes-Benz dealership group, where I implemented company-wide software systems to improve operations and drive revenue, and helped guide the group through a successful exit. Then enterprise SaaS at the customer-vendor seam, as a GTM strategy, ops, and internal solutions consultant. Now independent AI builds. Everything below is live. Every item links to a working tool, a deployed app, or a real artifact.

Agents Skills Products Workflows
Eight capabilities
01
Architecture Lead

OS Ingestion: a serverless edge-to-enterprise pipeline

The pattern behind modern business-process automation: unstructured edge data, a structured enterprise store, AI synthesis. A person talks into a phone; seconds later it's queryable knowledge. Zero database, zero hosting cost, lives entirely inside Google Workspace.

Live · Working · Tested Serverless Zero-cost infra
01 Mobile PWA Voice or text capture in the field
02 Apps Script webhook Serverless middleware, no backend to run
03 Google Doc data lake Append-only structured store
04 NotebookLM synthesis Source-grounded RAG layer over the lake

A React 18 PWA captures voice (Web Speech API) or text on a phone and fires it at a Google Apps Script webhook. The script appends each capture to a Google Doc acting as a data lake, which NotebookLM indexes as a retrieval layer. Talk on Monday, query the synthesized knowledge on Friday.

NotebookLM is Google's source-grounded research tool. It answers only from the documents you give it, which is what makes it safe as a synthesis layer over the data lake.

The same pattern transfers to anything that starts as a person talking and needs to end as queryable knowledge: client field notes, sales-call captures, inspection logs, customer voice intake, internal SOP queries. The plumbing is identical; only the front end changes.

This is the v1 of a Workspace-native operating system I can stand up for a client engagement in days, not months.

Stack

React 18 PWA Web Speech API Google Apps Script Google Docs NotebookLM

See it

Live app → Repository →
A captured note in the Google Doc data lake, timestamped and typed
NotebookLM answering a question grounded only in the captured note

A note captured on a phone, then queried and synthesized in NotebookLM.

02
Execution Lead

Wave 1: P&C broker outreach, in motion

I build the public-facing intelligence asset and the operational machine that uses it to source conversations, end to end. The same shape transfers to any vertical: build the deep-research artifact, wrap a contact pipeline around it, run the cadence.

Infrastructure live · list populating Sales motion
01 Apollo saved search Mid-market commercial P&C brokers
02 CRM pipeline Structured contact + stage tracking
03 Daily DM cadence 10 LinkedIn DMs/day from the queue
04 Reply runbook Documented handling, not improvised

A P&C broker briefing hub: a research-grade intelligence page built to open conversations, not to sell. It's the thing the outreach points to.

Behind it, an Apollo saved search feeds a CRM pipeline, a daily ten-DM LinkedIn cadence works the queue, and replies are handled against a written runbook so the motion is repeatable rather than dependent on me being clever each morning. Infrastructure is live; the Wave 1 list of 150 mid-market commercial P&C brokers is populating now.

Shape

Apollo CRM pipeline LinkedIn cadence Reply runbook

See it

Broker Briefing Hub →
03
Tooling Layer

The build process, productized

Three deployed apps. A thin layer on top of the AI tools that codifies my build process and makes it portable. The same thinking transfers to any client where the bottleneck is how to brief work to an AI tool.

Prompt structuring

ChatGPT Operator Trainer

Turns messy voice dumps into structured prompts across 12 frameworks: voice-dump cleanup, strategy-partner mode, build briefs for Codex and Claude Code, project setup, agent mission briefs, deep-research decision memos, product-signal extractors. The frameworks are a PRD methodology made interactive.

Open tool →
Workflow design

Codex Starter Kit

A browser-based agentic workflow builder. Rough task in, mission brief out: objective, deliverable, recommended skills and plugins, agent vs. subagent split, execution steps, verification standard, acceptance criteria. It's how a vague ask becomes something an agent can actually run.

Open tool →
Prompt transformation

Gemini Prompt Optimizer

A no-API, fully local prompt transformer for strategy, code, creative, data, and execution tasks. Built to need zero login and zero backend, the kind of friction-free tool you can hand a client without provisioning anything.

Open tool →
04
Intelligence & Research

Diagnosis, competitive research, client-ready packaging

Taking a business, researching its market, and packaging a clear-eyed recommendation a decision-maker can act on. Built with AI tools end to end: research, scoring, and the deliverable itself.

Vertical intelligence

P&C Broker Briefing Hub

The research-grade intelligence asset behind the Wave 1 motion (Section 02). Built to open conversations in the commercial insurance vertical.

View hub →
Industry audit

CDK Digital Retail Audit

A structured audit of digital-retail tooling in the automotive dealership space, the same diagnose-and-recommend pattern applied to a software category rather than a single business.

View audit →
Self-initiated diagnostic POC

Panozzo Bros.: digital audit and competitive research

A self-initiated, value-led diagnostic POC for a real 100-year-old family funeral home. I ran a full competitive digital audit, market research across the local competitors, and a client-ready set of recommendations, end to end with AI tools, on spec. The audit scorecard and the recommendations are linked below.

Audit & competitive research → Recommendations →
05
Agents & Agentic Systems

Working agents, not agent demos

Patterns that drive real APIs and real workflows. Each one transfers to a SaaS a client already runs.

Daily driver Customized environment

I run Claude Code as a customized agentic environment, not stock. Custom slash commands I built auto-load into every session (two are below). A claude-mem plugin carries memory across sessions. I direct Claude Code and Codex to build new agents, subagents, skills, commands, and plugins on demand, and I run automated code review inside my own build loop.

/boris

Loads Claude Code workflow patterns from the tool's creator, Boris Cherny, so every session starts with the best-practice playbook already in context: parallel sessions, plan mode, verification, subagents, hooks, and more.

/skill-router

Routes any task to the right skill or chain of skills from the live session catalog, with visible reasoning before it acts. Returns a numbered chain with a one-line rationale per skill, then asks whether to run it or just list it.

The /boris slash command and its description inside Claude Code
The /skill-router slash command and its description inside Claude Code

Two of my custom slash commands, auto-loaded into every Claude Code session.

Lead sourcing

SDR Agent

Python CLI: Google Maps lead source, website-quality scoring, ownership classification, Sheets export. Early build, working MVP, and the pattern that transferred straight into the Wave 1 broker pipeline.

Repository →
Multi-agent architecture

Auto Collision Intelligence

Paused

A five-agent system. Build complete; market entry paused due to former-employer adjacency. Included here purely as agentic-architecture proof, not a product I'm pushing.

View system →
API-driven CRUD

Notion Agent Hub

A request-file workflow plus Python CLI that drives Notion CRUD via API. The agentic pattern transfers to any SaaS with an API: GoHighLevel, HubSpot, Airtable, Salesforce.

Repository →
06
Products & Apps

Deployed apps

Smaller finished builds, live and usable.

Market research

Niche Validator

A tool for pressure-testing a niche or business idea before committing time to it. Deployed and usable.

Open app →
Personal utility

Workout Log

A clean, no-friction workout tracker. Small and finished.

Open app →
Communication tool

SignalSpeak

A communication and signal utility. Still very much an MVP.

Open app →
07
Training & Enablement

Bringing a team up to speed

The enablement side: onboarding material, lab work, and structured reporting that turns AI capability into something a team can adopt.

Productized enablement

Personalized AI Onboarding

Personalized AI onboarding built for one business at a time, delivered as a working web app, not a deck. The first edition is for private credit: five core principles, a copy-paste prompt library, an interactive 30-day program, and real deal workflows. The same framework rebuilds for any industry, down to a single SMB, and takes a non-technical team from skeptical to confident inside the tools they already use.

Open the app →
Build lab

Claude Code Lab

My working lab for Claude Code patterns: skills, agents, and workflows. The bench the rest of this portfolio is built on.

Repository →
Structured reporting

Anthropic Reports

Structured research and reporting work: turning model capability into something a decision-maker reads and acts on.

Repository →
How I work

Tool-agnostic by design

I pick whatever gets to a working result fastest, and I pick up new tools in an afternoon. No traditional engineering background. The range below is the modern AI build stack, plus the judgment for when to use which.

Build & ship
Claude Code, Codex, Cursor, VS Code, GitHub, Vercel
Strategy & reasoning
Claude Opus, ChatGPT, Gemini, Grok
Research
Perplexity, Gemini Deep Research, EXA, NotebookLM
Outreach
Custom MCP connectors and APIs, Instantly.ai
Data & ops layer
Google Workspace (Sheets, Apps Script, Drive)
Capture
Wispr Flow voice-to-text, the OS Ingestion PWA
Agentic layer
Custom Claude Code slash commands, the skills router, MCP connectors, and a Hermes agent (in build)
08
Operator Background

I have run a real business. The AI work sits on top of that.

Finance, then consulting at SullivanCotter, then operations leadership at a family Mercedes-Benz dealership group (grew 2 to 4 dealerships, $1,900 PVR above industry, led a successful exit to Fields), then Solutions Consultant at CCC Intelligent Solutions (Salesforce, Asana, and Power BI implementations, doubled NPS response rates, cut GTM cycles 25%). The AI work sits on top of all of it.

Career Work portfolio →

While not all-encompassing, as I'm learning and building every day, this is a show of capabilities and inventory.

Reach me here.