Pack

Deep Research System

Not a read-it-later pile. A pipeline that reads it and hands you the answer.

Python 3.10+ for the scripts, an LLM for the analysis step, and a Discord or Slack webhook if you want delivery.

// what it is

The raw material of competitive intelligence is everywhere — Reddit threads, HN posts, GitHub repos, blog essays — and turning that flood into something actionable is manual, slow, and inconsistent. You bookmark things you never revisit and have hunches you never validate.

This pack is the architecture for a pipeline that scans your sources continuously, triages incoming links by relevance and novelty and signal strength, extracts the content, analyzes it into actual intelligence rather than a summary, and delivers findings to Discord or Slack.

Five stdlib-only Python 3.10+ scripts (863 lines total) come with it, including select_top.py — the selection engine with keyword boosts, meme filtering, and a ranked shortlist. Reddit note carried from the pack: unauthenticated .json endpoints return 403 as of July 2026, so feed reddit_parse.py JSON you obtain another way; the HN, blog, and selection scripts are unaffected.

// why it works

25+ page guide covering scanning, triage, extraction, analysis, synthesis, and delivery
A scoring rubric for triage, not vibes — plus deep-dive and synthesis prompt templates
5 stdlib-only scripts, 863 lines, no dependencies and no install step
Worked examples of every output stage, so you know what good looks like