All Work

I built a Slack-based system for requesting and approving access across dozens of applications. It handles different approval paths, provisions approved access, and keeps an audit trail without turning every request into a security ticket.

Onboarding touched enough systems that copying the same logic between scripts stopped making sense. I turned it into a shared engine with dry runs, credential safeguards, device checks, and append-only audit logging.

A state-based workflow moves accounts through onboarding, first sign-in, and departure. It reduces manual cleanup while keeping destructive actions visible and controlled.

Automation can fail quietly, so I compare deactivated accounts across identity, email, and collaboration systems. The control reports discrepancies without trying to “helpfully” modify anything.

I automate the repetitive parts of Google Workspace administration without hiding what the automation is about to change. The work spans account lifecycle controls, directory cleanup, group management, and connected email-security systems, with dry runs and human review around destructive actions.

This automation keeps directory information synchronized with Proofpoint. Safe changes are additive and reversible; onboarding and offboarding actions remain human-reviewed.

I worked across endpoint and device-management data to find unhealthy installations and mismatched records. More than 100 records were corrected without misplacing a device.

A reconciliation uncovered that vendor and agreement records were living in too many places. I designed a normalized system that keeps the source record, review status, and compliance views connected.

What happens when a fun pet-photo tool accepts untrusted images and processes them on a server?

Memory safety, fuzzing, Django exploits, Kubernetes deployments, and Android privacy—with the messy debugging notes left intact.

Privacy rights should not require paying the same industry that profits from personal data. I objected to paying a subscription to remove information that should not have been collected in the first place, so I made the process free and kept the data in the browser.

I wanted a network-level way to block trackers and ads across the devices on my Wi-Fi, not just in one browser. A Raspberry Pi 4B running AdGuard Home as the DNS layer for my home network, blocking tracker and ad domains while preserving the services I actually use.

I have never experienced a day as a vertical list. Orbit started as an attempt to make time look more like it feels: circular, finite, and constantly being rearranged.

Some routines should reset from the last time they were done. I do not change my sheets every Tuesday—I change them a certain amount of time after I last changed them. A surprising number of routines work this way.

I kept opening world clocks and then doing the time-zone math myself anyway. This puts every city on one movable timeline so the overlap is visible.

When I was learning to code, I challenged myself to recreate a cat-snowflake meme in Java. That was the entire idea. Years later, I rebuilt it for the App Store because apparently I remained committed to the premise.

Group travel gets complicated fast when friends hold different passports. I started with a giant spreadsheet; eventually the spreadsheet became an argument for building an app.

Route planners regularly suggest trips that are technically faster and obviously worse. I wanted one that understood when I would rather walk farther, avoid a transfer, or take the ferry because the ferry is nicer.

Looking for a rent-stabilized apartment meant checking public records one building at a time. I wanted the useful parts on a map.

Catflakes needed isolated cat heads. I did not want to cut them out manually forever, so this became its own computer-vision project.

New Yorkers already have strong opinions about stations, neighborhoods, transit modes, public art, and pigeons. This turns those arguments into a deliberately impossible series of binary choices.

Most personality tests give you one tidy label and send you on your way. I wanted to see the messier version: overlapping traits, relative strengths, and changes over time.

I live near several cat colonies, encounter street cats almost everywhere I travel, and love cats enough to revisit a graduate-school model that was only right about 70% of the time.

I want to make my own wearable and take control of the data it collects.