Back

Second Brain

AI Agent

Second Brain is a personal AI system for research, product development, and building. It monitors market signals, validates product ideas against real evidence, generates specifications, and moves them toward working prototypes — compressing the path from insight to shipped experiment.

Intelligence & Research

Second Brain continuously monitors signals across AI, technology, and product. Every weekday morning an AI agent scans the web, synthesizes what matters, and delivers a structured briefing to my Obsidian vault. I get notified via Telegram when it's ready. Weekly and monthly summaries build on those daily signals, surfacing trends and product observations over time. Product signals come from X via Bird, which powers a weekly product briefing that mines builder threads and launches. Parallel powers the web search behind the briefings and on-demand research.

Everything I come across — articles, videos, GitHub repos, newsletters, X posts — gets saved to Readwise, my async intake pipeline. I save more than I can read, so a scheduled agent serves as my reading staff: it reads what I can't and compiles a personal knowledge wiki from those saves, following Andrej Karpathy's LLM wiki pattern — cross-referenced markdown pages an agent can query when I'm working on related topics. Second Brain also synthesizes the queue on demand, surfacing how new material connects to what I'm working on.

Validation

When something is worth pursuing, Second Brain runs a structured research and validation process. Using Last 30 Days, it searches Reddit and X for real pain points, existing solutions, and market gaps — grounding ideas in recent evidence before investing further.

Specification & Execution

Once a problem is validated, Second Brain generates high-fidelity product specifications and moves them toward working prototypes using Compound Engineering, Superpowers, and G-Stack for planning, brainstorming, and development within Claude Code.

Continuity Discipline

Every session ends with a continuity ritual — updating session-tasks.md with what got done and what's next, adding anything worth remembering across sessions to memory.md, and updating the global and per-project CLAUDE.md files when something changes about how I want Claude to work. The compounding effect: new sessions start with real context instead of a blank slate, and the system gets sharper over time rather than resetting.

Technology Stack

  • Mac Mini M4, always-on, home-hosted
  • MacBook Pro SSHs into the Mini over Tailscale; Claude Code in tmux so sessions persist across disconnects
  • NanoClaw (self-hosted agent framework, Docker, Node.js)
  • Claude via Anthropic API
  • Claude Code with Superpowers, G-Stack, and Compound Engineering
  • Bird, Last 30 Days, Readwise, Parallel
  • Obsidian vault synced to GitHub via automated git pipeline
  • Telegram bot for agent interaction