Work

6 studies

New business prospecting reporting across eight business lines

Executive briefing / synthesis

A VP spent 10 to 12 hours a month reconciling eight teams' numbers by hand. They now spend five minutes.

problem: Eight business lines reported prospecting performance in eight shapes. An executive consolidated them by hand each month.
approach: One normalization table that absorbs each team's existing output, validation at the point of entry, scripted assembly of the deck.
result: 10 to 12 hours of monthly VP effort down to a 5 to 10 minute review.
scale: Eight business lines. Fortune 500 automotive division. Reporting that reaches the enterprise CEO and the shareholder board.

read the study

One source, planned into every surface

AI enablement across two very different audiences

AI enablement / internal education

Most people in the division will never write code. A few will build applications. I run enablement for both.

problem: Everyone got a Copilot license and no reason to open it. Meanwhile the people ready to go further had nowhere to start.
approach: A monthly newsletter for the Copilot population, a Claude Code starter kit for the ones going further, and hands-on sessions in between.
signal: Global Automotive leadership now ranks among the heaviest Claude Code users in the enterprise.
scale: Global Automotive, a Fortune 500 division. Newsletter open to anyone at Assurant.

read the study

Evaluate, teach, adopt, repeat

A wiki that maintains itself

Knowledge management / AI automation

Program knowledge for a large organization lived in twenty places and decayed the moment anyone stopped maintaining it. I built a wiki that maintains itself: custom Claude Code skills scan the underlying corpus for gaps, redraft stale pages, and generate D3 visualizations — structured around how people actually search.

problem: Program knowledge for a large organization lived in twenty places and decayed the moment anyone stopped maintaining it.
approach: Custom Claude Code skills scan the underlying corpus for gaps, redraft stale pages, and generate D3 views of how the pieces connect — structured around how people actually search.
scale: Organization-wide corpus · refreshes against source, not memory

full study — request access

Self-maintaining knowledge loop

On-demand specialist routing in a regulated workflow

Zero-to-one product direction

A regulated transaction can't close without a licensed specialist, and scheduling one means customers wait while specialists sit idle. I designed a coordination model that routes the right specialist to each transaction in real time, with the compliance trail intact.

problem: A regulated transaction can't close without a licensed specialist, so customers wait while specialists sit idle somewhere else.
approach: A coordination model that routes the right specialist to each transaction in real time, with the compliance trail intact.
scale: Patent pending on the operating model

full study — request access

Constraint-routed assignment pipeline

A conversational assistant for frontline agents

Applied AI / customer-facing assistant

Frontline agents were losing the negotiation itself to product-detail lookups. I built a conversational assistant that surfaces the right specifics mid-call, and wrote the internal AI design playbook so the next team wouldn't have to re-derive prompt patterns from scratch.

problem: Frontline agents were losing the negotiation itself to product-detail lookups mid-call.
approach: A conversational assistant that surfaces the right specifics in the moment, plus the internal AI design playbook so the next team wouldn't re-derive prompt patterns.
scale: Piloted with agents; moving into production

full study — request access

Conversational loop with the agent in the middle

Plain-English answers from a hundred pages of governing documents

AI-augmented knowledge surface

My HOA's governing documents — a hundred-plus pages of CCRs, bylaws, and policies — were technically public but functionally unreadable. So I built a conversational assistant that answers neighbors' questions from those documents in plain English, with citations.

problem: My HOA's CCRs, bylaws, and policies were technically public and functionally unreadable.
approach: A conversational assistant that answers neighbors' questions from those documents in plain English, with citations back to the source.
scale: Side project · six weeks · real users

full study — request access

One document corpus, many answers

work

6 studies

New business prospecting reporting across eight business lines

Executive briefing / synthesis

A VP spent 10 to 12 hours a month reconciling eight teams' numbers by hand. They now spend five minutes.

problem: Eight business lines reported prospecting performance in eight shapes. An executive consolidated them by hand each month.
approach: One normalization table that absorbs each team's existing output, validation at the point of entry, scripted assembly of the deck.
result: 10 to 12 hours of monthly VP effort down to a 5 to 10 minute review.
scale: Eight business lines. Fortune 500 automotive division. Reporting that reaches the enterprise CEO and the shareholder board.

read the study

One source, planned into every surface

AI enablement across two very different audiences

AI enablement / internal education

Most people in the division will never write code. A few will build applications. I run enablement for both.

problem: Everyone got a Copilot license and no reason to open it. Meanwhile the people ready to go further had nowhere to start.
approach: A monthly newsletter for the Copilot population, a Claude Code starter kit for the ones going further, and hands-on sessions in between.
signal: Global Automotive leadership now ranks among the heaviest Claude Code users in the enterprise.
scale: Global Automotive, a Fortune 500 division. Newsletter open to anyone at Assurant.

read the study

Evaluate, teach, adopt, repeat

A wiki that maintains itself

Knowledge management / AI automation

problem: Program knowledge for a large organization lived in twenty places and decayed the moment anyone stopped maintaining it.
approach: Custom Claude Code skills scan the underlying corpus for gaps, redraft stale pages, and generate D3 views of how the pieces connect — structured around how people actually search.
scale: Organization-wide corpus · refreshes against source, not memory

full study — request access

Self-maintaining knowledge loop

On-demand specialist routing in a regulated workflow

Zero-to-one product direction

problem: A regulated transaction can't close without a licensed specialist, so customers wait while specialists sit idle somewhere else.
approach: A coordination model that routes the right specialist to each transaction in real time, with the compliance trail intact.
scale: Patent pending on the operating model

full study — request access

Constraint-routed assignment pipeline

A conversational assistant for frontline agents

Applied AI / customer-facing assistant

problem: Frontline agents were losing the negotiation itself to product-detail lookups mid-call.
approach: A conversational assistant that surfaces the right specifics in the moment, plus the internal AI design playbook so the next team wouldn't re-derive prompt patterns.
scale: Piloted with agents; moving into production

full study — request access

Conversational loop with the agent in the middle

Plain-English answers from a hundred pages of governing documents

AI-augmented knowledge surface

problem: My HOA's CCRs, bylaws, and policies were technically public and functionally unreadable.
approach: A conversational assistant that answers neighbors' questions from those documents in plain English, with citations back to the source.
scale: Side project · six weeks · real users

full study — request access

One document corpus, many answers