Building with AI

I treat AI as a tool for testing my own ideas rather than something that just hands me answers.

I have built a system that listens to a child talk about their summer research project and helps turn it into a report, and personal tools that handle my calendar and email in one place.

For a personal assistant using n8n and Discord, I integrated tools for Google Calendar, Tasks, Drive, Gmail, Contacts, Docs, and Sheets. It is a prototype for finding plans and organizing follow-up work from email. This implementation dates to July 2026; reconnecting services and checking startup after moving development to a Mac in September remained separate work.

Some failures had little to do with prompting. Adding an OAuth scope still required renewed consent, and an array schema missing its items definition caused an entire request to fail. I added a connection-check script so I could investigate authentication, API calls, and conversation separately.

Beyond convenience, I care that these tools never act on their own and that the person using them can see why something happened. Prototyping means constantly deciding where AI should act and where a human should decide.

In another operational tool, I made email approval cover the recipient, CC, body, and attachments. Changes after approval require another review instead of reusing the old approval. Making that relationship explicit in the implementation matters more to me than simply displaying an approved label.