Projects

AI-native systems I’ve shipped.

A working catalogue of the AI workflows, skills, and operator tools I’ve built and run in real revenue and personal-ops motions. Each one has receipts — time saved, outcomes hit, or a live link to the artifact itself.

01Live

Deal-Intelligence + Rep Coaching

A live product that reads your pipeline to flag deal risk and next-best actions, then coaches reps in the flow of the deal — the same AI-native playbook as the rest of this catalogue, shipped as something you can use today.

Live now at repcoaching.io

Deal intelligenceRep coachingSales AI

Visit Rep Coaching

02Demo

Apollo prospecting skill

Plain-English ICP → 50 ranked, enriched leads in under five minutes. Cuts a half-day prospecting workflow to one coffee. Live demo accepts your own Anthropic + Apollo keys.

4–6 hrs saved per ICP build

ClaudeApollo APIVercel Functions

Open demo

03Open source

GTM skill library

60+ open-source Claude Code skills for sales, marketing, and operator workflows. Drop-in skillpack for any Claude stack. Published v2-generalized branch.

Public skill catalogue, free to install

Claude CodeBashMDX

View on GitHub

04PHASE 0

DailyOK

Daily AI voice calls that check in on seniors and score eight dimensions of cognitive and emotional health from natural conversation — surfacing decline early for family and care teams. Caregiver dashboard, SMS alerts, HIPAA-grade data handling.

HIPAA-grade voice pipeline, pilot-ready

Voice AITwilioHIPAA

Read writeup

05FRAMEWORK

AI ROI Teardown

Turns an AI use case into the teardown a buyer actually needs — token and infra cost model, accuracy bar, failure-mode taxonomy, build-vs-buy math. The bridge between what engineers ship and what a CFO will sign.

Built as buyer + interview artifacts

ClaudeCost modelingEval design

Read a teardown

06FRAMEWORK

LLM Council

Routes a decision to five independent AI advisors, has them anonymously critique each other, then synthesizes a weighted verdict. Catches the blind spots a single model talks itself into. Modeled on Karpathy's LLM Council.

5 models, anonymous peer review

ClaudeMulti-modelPeer review

See the method

07AUTOMATION

Inbox Copilot

Reads important-unread mail, pulls deal and CRM context, and drafts replies matched to sender type — then learns from which drafts I send, edit, or kill. Drafts only; nothing leaves without me.

Learns from every edit

ClaudeGmail APILearning loop

Read the writeup

08AGENTS

Maintainer Orchestrator

A control-plane agent that triages multi-repo work, hands implementation to worker subagents, and brings back decision-ready PRs. Coordination and execution on separate planes — a real multi-agent system, not one long prompt pretending to be one.

Control plane + worker subagents

Claude CodeSubagentsGitHub

See the architecture

Want one of these wired into your stack?

Most of these are open-source or live demos. The rest I'm happy to walk through. I reply to every real message within 48 hours.

Get in touch Browse GitHub