🔍 I like digging deep and sharing what I learn · 🤖 These days I build together with AI
- 🧩 I build iOS apps and enjoy going deep on modularization and architecture
- 🛠️ When I need a tool, I build it myself — from developer CLIs to full products
- 🤖 I prototype fast with AI and turn ideas into shipped software
- ✍️ I like to write down and share what I learn along the way
Language & IDE
UI
Concurrency
Architecture
Web Bridge
Collaboration
Things I designed, built, and shipped on my own.
haetae-robotics — Safety & security for physical AI: model output is treated as untrusted input before it ever reaches a robot.
| Repo | Platform | Description |
|---|---|---|
| haetae |   |
A supervisory policy gate for AI-driven robots — every VLA / foundation-model action is allowed, slowed, or denied before it reaches the robot's own safety layer (pre-alpha) |
| ataxia | |
derailment for embodied agents — induces psychopathology-like behavioral distortions in robot policies in simulation and measures the degradation against a healthy baseline |
Static analysis tools that turn codebases and database schemas into a queryable dependency graph. Dead code, cycles, impact analysis, layering rules, and architecture metrics all become queries over one graph.
| Tool | Platform | Description |
|---|---|---|
| cartograph | |
Queryable dependency graph for Swift / iOS codebases, built on IndexStoreDB |
| kartograph |  |
Explainable dependency graphs and dead-code analysis for Kotlin / Android |
| gartograph |  |
Queryable dependency graph for Go codebases — cycles, dead code, and layer rules as queries |
| rustograph | |
Queryable dependency graph for Rust / Cargo workspaces — cycles, dead code, and impact analysis as queries |
| dartograph |  |
Evidence-backed dependency graph analysis for Dart / Flutter codebases |
| schemagraph | |
Dependency graphs for databases — catalogs and SQL bodies in, impact analysis, cycle detection, and architecture rules out |
Cross-language graphs · isthmus links the graphs above across language boundaries, fed by TypeScript and Python producers · click to expand
| Tool | Platform | Description |
|---|---|---|
| isthmus |  |
Joins the graphs above where per-language tools stop — Flutter MethodChannel/Pigeon (Dart ↔ Swift/Kotlin) and React Native bridge calls, HTTP client ↔ server routes, and code ↔ DB-schema persistence links |
| tsograph | |
Feeds isthmus from TypeScript / JavaScript (Node) services — HTTP routes (Next.js, Express, Fastify, NestJS…), OpenAPI contracts, ORM relation facts, and call graphs |
| pythograph | |
Feeds isthmus from Python services — Django / DRF / Flask routes, Django & SQLAlchemy relation facts, and call graphs; parses with ast only, never imports your code |
Tools for running AI coding agents safely and smoothly — sandboxing, policy gates, account switching, outcome accounting, and shared knowledge. Deterministic and local wherever possible.
| Tool | Platform | Description |
|---|---|---|
| agentbelt | |
Confines AI coding agents on macOS to a single project directory — Seatbelt-enforced, kernel-tested |
| exitzero | |
Offline policy and verification gates for AI-assisted coding teams |
| multi-account-tool | |
Switches between multiple AI CLI accounts (Claude Code, Codex, Gemini/Antigravity) from one TUI — Keychain-backed with automatic rollback |
| knowledger | |
Domain-scoped knowledge agreements for KBs, LLM wikis, and AI workflows — with a Hyperledger Fabric adapter |
More agent tools · 8 more — policies, skill stacks, session traces, and continuity, click to expand
| Tool | Platform | Description |
|---|---|---|
| joinery | |
The linker for agent skill stacks — extracts each skill's implicit CLI / MCP / client-feature dependencies, dry-fits them against your machine, then locks and compiles per-agent configs |
| riskgate | |
One YAML policy that decides what your coding agent may do — allow, prompt, or deny every tool call, portable across agent CLIs, zero dependencies |
| yield-audit | |
Outcome accounting for AI coding agents: what survived, what it cost, what was wasted — includes an AI rework-rate lens for phantom productivity |
| epitaph | |
A repo-scoped ledger of rejected agent attempts — structured tombstones for rolled-back patches, consulted before the next agent retries the same path |
| agent2perfetto | |
Converts Claude Code session logs into Perfetto trace JSON — view agent sessions as timelines, tool slices, and context / cost counters |
| packet-ask | |
Sends only a scrubbed packet — never the raw repo — to personal coding-subscription subagents for review and second opinions |
| relay-continuity | |
Local, privacy-first continuity for AI-assisted work — evidence-backed resume context without storing code, chats, or telemetry |
| understatus | |
A calm macOS statusline addon for Claude Code — live CPU, memory, battery, disk, network, and AI-session info, with selectable themes |
Experiments that deliberately break a model's reasoning — and measure exactly how it breaks.
| Project | Platform | Description |
|---|---|---|
| derailment | |
Induces psychopathology-like cognitive distortions in LLMs by manipulating attention, salience, valence, and arousal, then scores the result against a healthy baseline — emulation, not diagnosis |
Mobile engineering, data, infra & testing · 7 more, click to expand
📱 For iOS & mobile engineering
| Tool | Platform | Description |
|---|---|---|
| locus | |
Where every UI element lives in source — a static traceability map from iOS accessibility elements to Swift code; CLI + MCP server |
| strictmigrate | |
Delegates a Swift 6 strict-concurrency migration to agents — the compiler is the judge, a journal is the single source of truth; spans the KMP boundary |
| coroner | |
Local post-mortem triage for iOS telemetry (.ips + MetricKit) — symbolicates, clusters, and journals crashes by version, then exposes them to coding agents over MCP |
| reproof | |
Self-hosted mobile QA — records issues as video + actions + fixtures, replays them deterministically on Android / iOS, and re-verifies AI repair candidates against the approved recording |
🗄️ For data, infra & testing
| Tool | Platform | Description |
|---|---|---|
| vecdiff | |
Diffs two embedding-index snapshots for vector-DB migrations and index audits — local, numpy-only, CI-gateable |
| localqpu | |
A local emulator of the IBM Quantum Platform API — a "test mode" for code that calls a quantum cloud: no queues, no cost, and any failure you want to simulate |
| lychgate | |
A deterministic custodian for repos you stopped tending — merges provably boring Dependabot PRs and holds the rest at the gate; no LLM, no admin or workflow permissions |
Native macOS apps, built together with AI. Distributed at coden.kr.
| App | Description |
|---|---|
| Storage Medic | Find and clean up what's eating your disk space |
| Nits | Fine-tune display brightness and color temperature |
Side projects I've shipped · click to expand
| Service | Description |
|---|---|
| PickMeJobs | Job postings, aggregated in one place |
| Oneul Buffet | Map of buffets and company cafeterias |
| EVCar | EV charging station map |
| PyeonDeal | Convenience store deals and promotions |
| PolicyHanun | Government policies at a glance |
| Hansan | Real-time crowd density map of Seoul |
| Mosquito Index Map | Map of mosquito and pest activity |
| Observed | Real-time status monitoring for AI tools and infrastructure |
Thanks for stopping by! Feel free to reach out by email 😊