Index — Projects
Things I've
actually built.
Four projects — one long-running assistant, one productivity app I lived inside during exams, one gesture controller that escaped from Aria, and one tool for sales reps who talk faster than they can think.
01 — AI desktop assistant
Aria
A voice that listens and answers without feeling like a product.
ARIA started as a small idea and became the project I keep returning to. Months of rewrites, prototypes, and half-working modules that taught me how software actually holds together.
It's never been public. I never trusted the janky code enough. But some of the pieces were sharp enough to spin out on their own — the gesture controller below is one of them.
02 — Productivity, zero gaps
Wired In
A productivity app built around one idea — never leave a gap between tasks.
Two modes. No overlap. You set start and end times and the app makes sure you're always doing one of them. I built it to survive exam season and ended up using it every day.
Feel free to take the code and expand on the concept, or just make an account at wired1n.xyz and try it as it is.
Features
- Zero-gap focus, two modes
- Start / end times with persistent storage
- Progress tracking per mode
- Flip animation between modes
- Current task indicator with progress bar
- PWA — add to home screen on mobile
03 — Python · OpenCV · MediaPipe
Gesture Controller
Real-time hand gesture recognition, originally a limb of Aria.
MediaPipe tracks 21 landmarks on the hand and the program decides which fingers are extended to classify the gesture. Fist, index, two, three, open palm, thumbs up — recognised live on the webcam feed.
It started as one component of Aria before I split it into its own repo. Future ideas: custom gestures, swipe detection, desktop automation, keyboard shortcuts, multi-hand support.
Gestures
- ✊ Fist
- ☝️ Index finger
- ✌️ Two fingers
- 🤟 Three fingers
- 🖐 Open palm
- 👍 Thumbs up
04 — AI for live sales calls
ObjectionIQ
AI-powered objection handling for real-time sales conversations.
Reps pick or type an objection, add context, and get back a structured response — a reframe, a script, a follow-up question, and the pattern behind why the objection came up. The more data logged, the sharper it gets.
Early-stage, still moving. Connect your own API to try it. Nothing is stored server-side, and no keys live in the code or env — cap your spending limits if you want to be extra safe.
Features
- AI-generated response strategies
- Quick selection of common objections
- Custom objection input with context
- Structured outputs: reframe, script, follow-up, pattern
More on the way. If any of these are useful, take them, fork them, or send them somewhere better than where I left them, they're on my github.