Projects

DeFi has a front-running problem. High-frequency quant firms exploit transparency to jump in front of large trades — the little guy always loses. Arcane Pool is a dark pool AMM built natively on Avalanche using EERC (Encrypted ERC-20s), Avalanche's on-chain privacy primitive. Dark pools have existed in TradFi for decades; we brought them on-chain.

• Built for institutional and large-scale DeFi traders who need to move significant positions without being picked off by bots or MEV searchers.
• Because it runs on EERC, encryption is at the token level — trades settle on-chain with full verifiability but zero visibility until the moment you choose.

Every time you switch apps to ask an AI about something on your screen, you break focus and lose context. Prota started from frustration with that friction — the idea that AI should be ambient, not another tab you switch to.

• Understand whatever is on your screen — an error, a document, an email — and get an AI response without ever leaving the app you're in.
• Designed to feel invisible: insights surface in your notification shade, in and out in seconds, then gone.

I wanted to understand what a neural network actually is at the lowest level — not a PyTorch call, not a matrix abstraction, but actual logic gates switching. So I built one in hardware. No OS, no drivers, no CPU; just Verilog running at clock speed.

• Proves that ML inference doesn't need a cloud GPU — useful for always-on edge devices where power and latency budget is tiny.
• A forcing function for understanding ML from first principles: when you're writing the multiply-accumulate in gates, there's nowhere to hide.

Every time you open a new chat — switching from Claude to GPT to Gemini, or starting a fresh session — you re-paste the same background. It's mindless repetition that adds up. Context Folder stores your context once and injects it wherever you need it.

• Useful for anyone juggling multiple AI tools: drop your project brief, codebase context, or team guidelines in once and reuse everywhere.
• Especially built for teams — share a context folder so everyone's prompts start from the same page.

Attention spans are shrinking — and short-form platforms built on intrusive, algorithmic content are a big reason why. We built Knowble because we believe the format itself isn't the problem; the content is. Knowble is an AI-powered short-form learning platform where only educational and learning videos exist.

• No entertainment rabbit holes, no viral noise — every scroll surfaces something you can actually learn from.
• AI lets you go deeper on any clip: ask questions, get explanations, and turn passive watching into active understanding.

Laundry is one of those chores that every generation assumes automation is right around the corner — and yet here we are. This started as a semester project and turned into a genuine attempt at solving a hard robotics problem: making a machine that handles soft, unpredictable material reliably.

• A real physical machine, not a simulation — it irons, it folds, it works. Built and tested with actual fabric.
• The interesting challenge wasn't the mechanism; it was orchestrating three independent subsystems to hand off to each other without a hitch.

I wanted to know what it felt like to build something you could actually play — entirely in hardware. No game engine, no operating system, no abstraction layer. Just Verilog, a VGA cable, and the Basys 3 sitting on a desk.

• Everything from the random car positions to the collision detection runs as hardware logic — it's the most direct relationship between code and screen you can have.
• An honest excuse to learn digital design through something fun, because finite state machines are a lot more compelling when the output is a moving car on a screen.

Tools like Illustrator feel like magic until you try to build one. This was a deliberate exercise in demystifying graphics software — understanding how undo/redo actually works, how SVG is parsed, how a canvas tracks and redraws state. Built it to understand it.

• A fully working SVG editor you can use for real lightweight design work without pulling in a gigabyte of Electron.
• The deeper goal: if you can build the tool, you understand what every other tool in the category is doing.