What I build when nobody assigned the problem. Each one gets the same write-up: the problem, why I took it on, the value, and how I solved it. I'll keep adding to this page. Hours are my own estimates.
Showing all 7 projects. Hours are my estimates.
BW-1Web build
Ongoing
This personal portfolio site
Under 10 hours so far (my estimate)
Problem
I wanted one place that shows what I do across AI enablement, product ownership, project and program management, and Salesforce.
Why
It is also practice at planning, testing and running something real with AI tools, from my side as a non-engineer.
Value
A live site with automated checks and tight security settings that I can keep adding to. It uses no cookies and no analytics.
How I solved it
Plan each change first, then run the tests before anything goes live.
Harden the security headers and the free-plan Cloudflare settings.
A daily scheduled check confirms the live site still passes.
BW-2Agent coaching
Ongoing
Building AI agents and skills with friends and family
About 10 to 30 hours so far (my estimate)
Problem
People I know in healthcare, finance, manufacturing and tech have a daily task that eats their time, and they don't know how to hand it to an AI tool.
Why
Agents aren't only for IT. A good one starts from a real frustration, not from a tool.
Value
Each person ends up with an agent or skill that fits their own work, and they know how to adjust it themselves. I haven't measured time saved. Names and details stay private.
How I solved it
Interview first, so the agent matches how they actually work.
Write test scenarios and run them before they rely on it.
Work inside the tools they already have, and teach them to change the instructions on their own.
BW-3Jira app
In progress
Stratus for Jira
About 10 to 20 hours so far (my estimate)
Problem
Salesforce release updates and security risks come with deadlines, and administrators end up transcribing them into tickets by hand.
Why
I know this problem from the admin side, and it is a good chance to build a real app on the Atlassian platform.
Value
When finished, administrators no longer transcribe deadlines by hand. It runs in a personal test environment and is not yet listed on the Atlassian Marketplace.
How I solved it
It previews what it will create, then opens one ticket per issue with a due date and a plain-language summary.
It updates due dates when Salesforce changes and never overwrites a ticket a person has edited or closed.
The product brief, requirements and backlog are mine, and I built it with AI coding tools.
BW-4Home tech
Complete
Smart TV cleanup
Under 10 hours (my estimate)
Problem
The family room TV was full of ads, tracking and demo apps, and the home screen pushed content I didn't pick.
Why
I wanted a TV that opens to what we watch and nothing else, without breaking anything the household uses.
Value
The TV is quiet and fast to use, and every change can be undone. A second device got the same treatment afterward.
How I solved it
Back up each device first and keep an undo record.
Switch apps off instead of deleting them, and keep streaming, casting and the voice remote working.
Write a how-to so the same steps work on the next device.
Result 27 apps switched off on the TCL TV. My count.
BW-5Claude mod
Complete
Claude model-router mod
Under 10 hours (my estimate)
Problem
Claude Code helpers that only search and read were running on the same model as everything else.
Why
Simple lookups don't need a bigger model, and I wanted to see where helper work goes.
Value
Cheap lookups go to a cheaper model by default, and I can still pick a model on purpose. I haven't measured the savings.
How I solved it
Send lookup-only helpers to Haiku and other helpers to Sonnet.
A model I name myself always wins.
Keep a running count of helpers per model in the status line.
BW-6Data and automation
Ongoing
Fantasy football analysis tools
Under 10 hours so far (my estimate)
Problem
Running two fantasy leagues well means checking rosters, injuries, weather, waivers and trade offers every week.
Why
It's a low-stakes place to practice the same habits I use at work: automate the routine, keep a human on decisions.
Value
A consistent weekly routine instead of ad hoc research. Trade offers get a clear rating with reasons. Whether it wins more is not measured.
How I solved it
Pull league data into local files at the start of each session, then analyze rosters, free agents and trades.
Check injuries and weather before lineups are set.
Any message to another team owner waits for my approval.
BW-7Agent design
In design
Email-in-your-voice agent
Under 10 hours (my estimate)
Problem
Someone I know spends a lot of time writing routine emails and wanted a Microsoft 365 Copilot agent that writes them the way they do.
Why
I interviewed first because the agent has to sound like that person, not like a generic template.
Value
Both interview rounds are done and the agent instructions are ready. It hasn't been tested in Copilot yet, so I can't say how well it works.
How I solved it
Two rounds of interview questions to capture their voice and rules.
Write the agent instructions, a voice profile and starter prompts.
Prepare test scenarios and a scoring sheet for the first real test.