Portfolio Details - Site Agents

LLM-agnostic AI agents that audit, plan, and maintain portfolio websites autonomously through a Notion + GitHub workflow.

Notion backlog showing Planner agent tasks with rationale and acceptance criteria
Example GitHub pull request opened by the Dev agent with structured description
Agent architecture diagram showing Planner and Dev agents sharing state through Notion

Project information

  • Category: Machine Learning and Artificial Intelligence, Scripting and Automations
  • Client: Internal Project
  • Project date: January 2025
  • Status: Live · in active internal use
  • Stacks: Anthropic Claude (via OpenAI-compatible API), Notion API, GitHub API, Google PageSpeed Insights API, Node.js

Site Agents — AI-Powered Portfolio Management

Two LLM-agnostic AI agents that manage my portfolio websites. A Planner agent audits the live sites (SEO, performance via PageSpeed, Open Graph, images) and files tasks in a Notion backlog with rationale and acceptance criteria. A Dev agent (OpenCode) picks up approved tasks, writes the code on a branch, and opens a GitHub pull request. I approve tasks and merge PRs; a GitHub Action deploys to Hostinger. The agents share state through Notion, keep long-term memory in markdown files, and a retro step turns Dev's questions into better planning rules.

Key Features

  • Automated SEO and performance audits using Lighthouse/PageSpeed
  • Task planning with concrete acceptance criteria in Notion
  • Code generation following site-specific conventions
  • GitHub PR workflow with human approval gates
  • Multi-site monitoring (visio, tech, vue portfolios)
  • Long-term memory via markdown files for learning preferences
  • Retrospective loop that improves planning from dev feedback