Projects
Building things to understand how they work. From research papers to systems programming experiments.
ZeroKey
A distributed execution model separating instruction, execution, and accountability in blockchain systems. Research paper with working implementation.
Key Highlights
- Novel separation of instruction and execution
- Threshold cryptography (t-of-n) signing
- Formal model with security proofs
- Research paper ready for publication
Koto
「あれやっといて」を、次の入力候補に変える AI Native IME。A local-first AI Native IME that turns rough intent into context-aware input candidates — reframing input from reading-to-text to intent-to-text, while suggesting (never acting) under user control.
Key Highlights
- Defines a new category: AI Native IME
- Intent-to-Text Transformation from local working context
- `Suggest, not act` — user stays in control
- Position paper published
TenkaCloud
BULL's open-source, multi-tenant platform for running cloud challenges, GameDays, and technical learning events. It is becoming the next pillar of BULL alongside our engineering engagements, and reads as proof of implementation for multi-tenant operations, identity and access, and auditability — the exact problems we solve for clients.
Key Highlights
- Multi-tenant control-plane / application-plane architecture
- Role-based access and SSO for organizers, staff, and participants
- Admin controls and audit trails for accountability
- Operational tooling to run live Battle / Challenge events
Open Source Contributions
Learning & Experiments
Browser from Scratch
LearningBuilding a web browser from scratch in Rust. Deep dive into rendering engines, HTML/CSS parsing, networking, and modern web standards.
OS from Scratch
LearningImplementing an operating system in Rust. Covers bootloader, memory management, multitasking, interrupt handling, and device drivers.
Blockchain in Zig
LearningBuilding a blockchain from scratch in Zig, now written up as a free book. Covers blocks and PoW, P2P networking with gossip-based chain sync (verified across 3 nodes), and a simplified EVM that runs a subset of Solidity opcodes and ABI calls. A hands-on toolkit for learning by implementing — PoS is intentionally left as an open exercise.
DeepForm
ActiveAI-powered depth interview tool that transforms vague product ideas into production-ready specifications. Bridges non-technical stakeholders and engineering teams with evidence-traced requirements.
GitHub Vector Issue Processor
ActiveAI-powered tool that processes GitHub issues using vector embeddings. Enables semantic search, similarity matching, and intelligent issue triage.
annai.nvim
ActiveAsk your Neovim, in plain language, “how do I…?” — an on-device LLM (Apple Intelligence, with an Ollama fallback) answers with a real keybinding picked from your own keymaps. No cloud, no API key, no telemetry.