AI orchestration developer
grazomarin
Background visual: a trace of orchestration runs — a feature shipped through a six-way subagent fan-out, context assembly, cross-model review, model judging, memory distillation.
> traverse_constellation()
Projects
Every star is a real project; every line is lineage — solid where one piece of work directly grew into the next, dashed where the idea carried. Hover to trace a thread, select a star to read it.
Edge types borrowed from Graphify: extracted / inferred
- Web Foundations (2022, Foundations): Where it started — The Odin Project era. A year of vanilla JavaScript: a test-driven battleship, an async weather app, DOM games, forms, and a calculator. No frameworks, no shortcuts.
- Graphs & Trees (2022-23, Foundations): CS fundamentals in public: linked lists, binary search trees, and BFS pathfinding with Knight's Travails — later rebuilt as a TypeScript visualizer. The first graph thinking that resurfaces, years later, in Graphify.
- vscode-spd-handin (2023, Developer Experience): A VS Code extension that automated Racket assignment hand-ins over SSL for UBC coursework — the first build-a-tool-so-the-workflow-disappears project.
- AI Wardrobe (2024, AI Orchestration): First AI-powered build — a wardrobe assistant experimenting with prompt-driven outfit logic. The moment the front-end path bent toward AI.
- UBC AI Club (2024, Front-End Systems): Site work for the UBC AI Club — building for the campus AI community while studying it.
- WCAG Guidelines (2024, Front-End Systems): A TypeScript project around the WCAG guidelines — the accessibility groundwork that later shows up as the DatePicker's keyboard model and ARIA discipline.
- Build & Quality Infra (2025, Developer Experience): Performance, CI parity, and stability across enterprise front-ends at Samsung R&D — the unglamorous work that makes everything else fast.
- Component Library Work (2025-26, Front-End Systems): Design-system components and typed utilities for Samsung's shared component library — Toast, StoryCard, StoryCarousel, Selectable Card, and generics-driven hooks.
- DatePicker System (2026, Front-End Systems): A production DatePicker that survives edge cases: segment-based input, a full keyboard model, timezone correctness on month boundaries, and localization across 8 languages.
- Memory Bank & Cline Workflows (2026, AI Orchestration): A two-tier context system that turns a stateless AI assistant into a persistent collaborator — designed and shipped for Samsung's in-house Cline fork.
- Graphify Integration (2026, AI Infrastructure): Adapted the open-source Graphify knowledge-graph tool to a large Samsung monorepo — package-level graphs, resumable extraction, and git-hook automation that give an AI assistant durable structural memory of the codebase.
- gemini-cli-as-mcp (2026, AI Infrastructure): Adapted fork: an MCP server that exposes Google's Gemini CLI to other AI assistants — cross-assistant orchestration, one model calling another as a tool.
- grazomarin — this site (2026, Front-End Systems): The site you're looking at. An R3F galaxy with live GitHub-fed stats, a typewriter identity, and this constellation — built end-to-end in an agentic Claude Code workflow with specs, subagents, and review gates.
> inspect --node graphify
Graphify Integration
AI Infrastructure2026
Adapted the open-source Graphify knowledge-graph tool to a large Samsung monorepo — package-level graphs, resumable extraction, and git-hook automation that give an AI assistant durable structural memory of the codebase.
- Package-level graphs instead of one repo-root graph
- Sequential, resumable chunk-by-chunk extraction (no background workers)
- Tracked git hooks rebuild only the affected package graphs per commit
- pnpm graphify:setup / graphify:doctor one-command onboarding
- Pre-merge audit caught and resolved a Memory Bank policy conflict
- Edges labeled EXTRACTED / INFERRED / AMBIGUOUS — the convention this constellation borrows
Knowledge graphsGit hooksShellNodeCline
Samsung R&D · private repo
> man orchestrate_intelligence
The loop
The trace in the hero is not a mock; it replays how I actually ship. By day I build front-end systems at Samsung R&D. The rest, the agents, the memory, the review pipelines, exists because I wanted shipping with AI to feel as disciplined as shipping without it.
Every run walks the same five stages.
assemble_context()
A run starts by remembering. The distilled rules of past sessions load first, then the repo itself: structure, history, the graph of what touched what. No run starts from zero.
- curated memory · 30+ distilled rules
- Graphify graph
write_plan()
Nothing is built from a vibe. An idea is brainstormed into a spec, the spec into a plan, the plan into tasks small enough to review honestly.
- brainstorm → spec → plan, written before code
fan_out()
Implementation goes to fresh subagents, one per task, each handed exactly the context it needs and nothing more. Isolation keeps them focused; the orchestrator keeps the thread.
- subagent-driven development
- fresh context per task
scrutinize()
Every change faces reviewers that did not write it: spec compliance first, then code quality, then a second model with different instincts. A finding only counts once it is verified against the source.
- parallel-pr-review
- dual-review: codex → claude
- judge_panel: gemini · codex · claude
distill()
A session ends by writing back what it learned. Corrections become durable rules, decisions become memory, and the next run assembles a smarter context than this one had.
- session distillation ritual
- write-back → memory index
the next run starts smarter
> open_channel()
Send a signal
If you're hiring for this kind of work, or building with agents and want to compare notes, the channel is open. Email gets a reply; GitHub is where the practice lives in public.
session persisted · memory updated · exit 0