ibtk extract-messages.
ibtk grew out of a one-off investigation into a single iPhone backup. Generalizing that work meant turning it into a reusable tool with a subcommand for each data type.
An iOS backup does not store files under their original names. Files sit in a flat, hash-named structure, and an index database (Manifest.db) maps each hashed name back to a logical domain and path. Messages, contacts, calendars, and bookmarks then live inside SQLite databases of their own.
Any extractor therefore has to:
Skipping any of these gives output that looks plausible but is wrong: missing attachments, or dates decades off.
ibtk works on unencrypted backups that you own. It does not defeat backup encryption. It is for recovering your own data, such as photos from a dead phone or a message thread you need to preserve.
đź”— Repository on GitHub
]]>Photo Counter is a book compiled from daily ethnographic notes taken while working the photo counter at a Walgreens. The notes are recorded as daily field observations, with the aim of compiling them into a published book.
Subtitle under consideration: Notes from a Dead Zone.
]]>Analyzing tempo and key once, at download time, means that information is already in the file when it is time to make music.
The same philosophy drives Logic Bounce Sync: remove manual file management from the studio workflow.
]]>Gymneye is a consumer booking app that routes business travelers to vetted commercial gyms near where they are staying. It started as a hotel-gym-access idea and was repositioned toward routing travelers to vetted commercial gyms.
The plan depends on near-zero customer acquisition cost through corporate travel channels such as Navan, Concur, and travel management companies. Business travelers are already booking through those tools, so Gymneye plugs into the booking moment instead of competing for attention.
]]>I didn’t research this market. I’ve lived in it my whole life.
Limerance is a nine-track concept album following a rootless man caught in an obsessive emotional loop.
The narrative sits alongside The Great Gatsby, (500) Days of Summer, and The Sun Also Rises. The closest poetic parallel is Shakespeare’s Sonnet 147. A living album bible tracks the concept, motifs, and track-by-track structure.
Produced in Logic Pro under the alias drwavy.
]]>Languages / technology used: Python, Django, PostgreSQL, Railway / Render
SampleKit (working title TitledLake) is a native macOS sample browser for music producers. Every ML step runs locally on Apple Silicon, and nothing leaves the machine.
SampleKit is a Swift Package Manager project split into targets with clear boundaries:
cblas_sgemm from Accelerate. With normalized vectors, cosine similarity reduces to a dot product, so ranking a whole library is one BLAS call over a mapped buffer, with no per-row loop and no copy into memory.The package structure, storage layer, and benchmark harness are in place. The audio input path is the current open item: it needs the device.audio-input entitlement before AudioIO is implemented.
Could current headsets track people persistently and overlay information on them? I looked at Apple Vision Pro (M5, visionOS 26 and 27) and Meta Ray-Ban smart glasses.
Both are blocked by platform privacy policy, not hardware limits. The silicon can run person-tracking models. The operating systems and developer terms deliberately do not allow apps to do this.
That matters for the design. A product built on a capability the platform forbids has a hard dependency on a rule it cannot change.
Two viable directions came out of the research:
Both keep the validation problem and drop the person-tracking constraint.
]]>RAR is proprietary and other formats arent so its making a point by not using it. I shouldn’t have to dig through Homebrew documentation or fall down Reddit rabbit holes just to open a rar file though. I could install unrar through the terminal and call it a day, but I didn’t want to open terminal every single time I want to open a file. I should be able to click on it and it opens. I would like to experience the computer like a normal person. So I automated the whole thing.
Before I could automate anything, I needed the extraction tool itself. Homebrew made this pretty straightforward:
brew install unrar
Once installed, I had to figure out exactly where it lived on the system. This is important with macOS automation because their tools for some reason don’t always know where your stuff is.
which unrar
Output: /usr/local/bin/unrar (or /opt/homebrew/bin/unrar)
My first move was to build a macOS Quick Action. This would let me right-click a file in Finder and run a custom command from the menu. I threw together a Shell Script workflow with pretty straightforward logic: grab the selected file, find its directory, and run unrar.
for f in "$@"
do
cd "$(dirname "$f")"
/usr/local/bin/unrar x "$f"
done
Notice I couldn’t just write unrar x "$f" and be done with it.
When Automator (or Shortcuts) runs a shell script, it doesn’t load your user
profile. No .zshrc, no .bash_profile, nothing. So it has zero clue where
Homebrew puts things. I had to hardcode the full path: /usr/local/bin/unrar.
This worked, but it felt dated. Sharing Automator workflows is a pain. You have to physically send the .workflow file, the recipient has to install it in some specific library folder, there’s signing, there’s permissions. I wanted something cleaner.
I decided to rebuild everything in Shortcuts. Better UI, easier sharing (just an iCloud link), and it syncs across all my Macs automatically.
The core logic stayed the same, but the implementation felt way cleaner.
I created a new Shortcut, enabled it for Quick Actions in Finder, and used the Run Shell Script action with /bin/zsh as the shell.
Same as before, I had to be explicit with the file path:
for f in "$@"
do
cd "$(dirname "$f")"
/usr/local/bin/unrar x "$f"
done
Now when I download a RAR file, I don’t touch the terminal at all. Right-click, hit “Unrar Here,” and the folder just appears. I uploaded the shortcut to RoutineHub so I can manage updates and versions without any hassle.
]]>Part of my work has involved me using skills learnt from a previous personal project SMDDMDWRITER to build a custom CLI tool to streamline NDJSON metadata management.
Languages / Technology used: Python, JavaScript, Sanity.io, NDJSON
Project is still in-progress.
]]>