01
Why Chrome extension UI automation becomes whack-a-mole
Notes from building AI Studio Prompt Library: scoped MV3 messaging, layered DOM selectors, Angular form events, and user-controlled selector fallbacks.
I build practical AI tools,
local-first browser systems,
and repo-native developer workflows.
SYSTEM NOTE
Operational simplicity compounds.
SYSTEM NOTE
DOM behavior matters more than framework choice.
SYSTEM NOTE
Boring infrastructure beats clever infrastructure.
SYSTEM NOTE
Agent failures usually start with state drift.
/ Writing
Deep dives, Architecture decisions, implementation tradeoffs, and the engineering reality behind the systems I build.
01
Notes from building AI Studio Prompt Library: scoped MV3 messaging, layered DOM selectors, Angular form events, and user-controlled selector fallbacks.
02
How PromptReady moved past generic page cleanup by making local Markdown capture, source metadata, fallback selection, and AI cleanup boundaries explicit.
03
Notes on building Satori around deterministic MCP workflows, exact reads, stale-index warnings, and bounded retrieval contracts.
/ Domains
Systems I actively work on, grouped by problem area.
Repo tooling, MCP workflows, semantic retrieval, context infrastructure for coding agents.
Chrome extensions, page extraction systems, offline-first capture, and local Markdown pipelines.
Frontend architecture, dashboard interfaces, state management, and automation systems.
Crypto pair research CLIs, statistical validation workflows, and paper simulation tools.
/ Projects
A few practical tools, experiments, and case studies from my recent work in retrieval, browser tooling, and quantitative validation.
Project 01
search ──➔ retrieve ──➔ readAgent-safe semantic code retrieval for MCP coding agents.
AI coding agents often edit from partial repo context. Satori helps them search semantically, inspect symbols, follow call paths, and verify fresher context before editing.
Project 02
html ────➔ markdownOffline-first Chrome extension for rendered-page capture into clean Markdown.
Captures useful page content from messy web pages and turns it into prompt-ready Markdown for notes, research, and AI workflows.
Project 03
discover ──➔ validate ──➔ vetoA crypto pair-testing CLI that asks which signals deserve trust.
Compares spot pairs, checks whether the data and setup are reliable, uses advanced dependency modeling to test pair relationships, and exports paper-run decision reports before any external execution path.
/ Experience
A compact view of roles and product areas. Full details live on the resume page.
Audited B2B client applications to resolve interface latency and process friction. Engineered Satori MCP retrieval engine and PromptReady browser extension.
Led React/Angular frontend architecture and WORKAI AI product delivery. Built custom Angular canvas e-signature canvas engines to bypass third-party library constraints.
Designed wallet-connected swap/liquidity interfaces. Built React state machines managing multi-provider transaction approvals and RPC recoveries.
Developed high-density tax filing form validators. Built a Node.js Puppeteer scraping pipeline migrating 1,500+ rich-text articles in 1 month.
Designed Figma wireframes and built responsive Angular client interfaces paired with custom automation scripts.
/ Stack
Grouped by how I use the tools, not as a logo wall.
React / Angular / TypeScript / WXT / Chrome MV3 / Vite / RxJS / NgRx / Redux
Node.js / Express / Hono / PostgreSQL / Supabase / Milvus/Zilliz
MCP / semantic retrieval / offline Markdown capture / OpenAI / OpenRouter / VoyageAI / Puppeteer
Python / Typer / copula models / pair validation / paper-run reports
ethers.js / wagmi / thirdweb / 0x
Field note archive
I use this site to publish the systems, tradeoffs, experiments, implementation notes, and engineering decisions behind the tools I build.