Project story

Agentic Portfolio

Production AI products with Apple-level engineering quality.

10 WebMCP tools
100/100/100/100 Lighthouse CI gate
<50ms Local actions
15 Monitor sections

Demo

Walkthrough on YouTube

Problem

Recruiters need to understand a builder in 30 seconds, but most portfolios are static resumes with stock themes and no proof of production AI craft.

Why existing solutions fail

Template sites hide architecture, chatbots are generic wrappers, and GitHub grids do not tell a product story or show operational maturity.

My approach

Dual-host static SPA + FastAPI on Vercel, local-first WebMCP for instant actions, OpenRouter only when site knowledge cannot answer, public monitor for ops proof.

Tradeoffs

GH Pages visitors pay a cross-origin API hop; local regex/tools handle navigation so LLM latency does not block core flows.

Results

  • 10 deterministic WebMCP tools — navigate, filter projects, download resume, toggle theme
  • Lighthouse 100/100/100/100 deploy CI gate on built dist
  • Public System Monitor with 15 diagnostic sections
  • 12 technical blog posts with structured “Things I Learned”
  • WWDC26 liquid glass across home, travel, monitor, blog, systems
  • Per-project case studies with architecture diagrams and evidence rows

What I would improve

  • Eval harness for AssistMe tool-calling accuracy
  • Dedicated 90s product walkthrough video per flagship project
  • Expand OSS line to MCP server + eval harness repos
FastAPIOpenRouteresbuildPlaywrightWebMCPVercelGitHub Pages