brain
projects·project·shipped

Portfolio OS

Thesis

"A self-evolving personal operating system that combines Notion-style content management with AI-powered features." The portfolio site itself — portfoliov2-ashen.vercel.app — but framed as an OS: structured content (MDX), a visual knowledge graph (D3), embedding-based semantic search, a streaming AI companion, and an autonomous reasoning agent that generates insights.

Status

shipped (live at portfoliov2-ashen.vercel.app). Created 2026-03-29, last push 2026-04-03.

Stack

  • Framework: Next.js 14 App Router, TypeScript, Tailwind CSS.
  • Content: MDX for project documentation.
  • Knowledge graph: D3.js visualization of project relationships.
  • AI: OpenAI API for embeddings and chat.
  • Scripts: npm run embed (generate content embeddings), npm run agent (autonomous reasoning agent for "memory layer").
  • Env: OPENAI_API_KEY, GITHUB_USERNAME, GITHUB_TOKEN (for repo fetching — the portfolio likely pulls live repo data).

Key decisions

  • MDX over CMS: project docs are markdown-with-components, authored as code. Reviewable, versionable, no separate content DB.
  • D3 for project relationships: visual graph of how projects connect is a first-class UI element, not a nav afterthought.
  • Embeddings + chat as native surfaces: semantic search and an AI companion are built in, not bolted on — consistent with Zaid's broader "AI-native not AI-bolted-on" stance.
  • "Memory layer" via autonomous agent: npm run agent implies a background/offline process that surfaces insights from the content — the portfolio isn't just static, it produces commentary.

Learnings

  • This is an AI-native artifact about AI-native work: portfolio-os uses embeddings and agents to talk about projects that themselves use embeddings and agents. Meta-coherent.
  • GitHub integration (GITHUB_TOKEN) suggests repos feed into the site — the portfolio stays fresh as Zaid ships.

Outcomes

  • Live portfolio site; canonical public-facing view of every project in this wiki.

Open questions

  • Does the "memory layer" agent actually publish insights anywhere, or is it experimental?
  • Is the knowledge graph auto-built from project frontmatter / tags, or hand-curated?
  • Is the private LLM wiki (this wiki) the eventual feeder into a public Portfolio OS view?

Links