Skip to content

raven-sourav/pmm-content-engine

Repository files navigation

Content Distribution Engine — AllAboutPMM

A multi-format content distribution engine for PMM leaders. Powered by a synthesized PMM Brain built from 13 newsletters (861 raw posts), 37 mental models, and 5 expert technique libraries. Produces evidence-backed, operator-credible content across LinkedIn, Twitter/X, Newsletter, and Carousel — not generic AI fluff.

How It Works

13 Newsletters → 861 Raw Posts → PMM Brain (37 models, 69 stats)
                                       ↓
                              Theme + Research
                                       ↓
                         ┌─────────────────────────────┐
                         │  3-Angle Generation Engine   │
                         │  (Contrarian / Framework /   │
                         │   Story-Led × 5 Hook Types)  │
                         └─────────────┬───────────────┘
                                       ↓
                         ┌─────────────────────────────┐
                         │   Rubber Duck Escalator      │
                         │   5-Phase Quality Gate       │
                         │   (Threshold: 8+/10)         │
                         └─────────────┬───────────────┘
                                       ↓
              ┌────────────┬───────────┼───────────┬────────────┐
              ↓            ↓           ↓           ↓            ↓
          Newsletter    LinkedIn    Twitter/X   Carousel    Visual
         (1500-3000   (800-1200    (7-10       (5-10       (Pillow
          words)       chars)      tweets)     slides)      image)

The Pipeline

  1. PMM Brain — Synthesized intelligence from 13 newsletters + 2 writer DNA profiles. 37 mental models, 69 evidence stats, 11 topic depth layers. This is the thinking engine.
  2. Research — Deep web research on any theme. Produces a Research Brief + Gap Map.
  3. 3-Angle Generation — Every theme produces 3 drafts: contrarian reframe (drives comments), framework/mental model (drives saves/shares), story-led with earned insight (drives reactions). Works across all 4 formats.
  4. Rubber Duck Escalator — 5-phase critique (MIRROR, PROBE, CHALLENGE, ILLUMINATE, CRYSTALLIZE) with format-adapted scoring. Auto-rewrites until quality threshold met.
  5. Distribution — Newsletter article is the pillar. Atomizes into LinkedIn post + Twitter thread + Carousel. Also supports reverse (expand LinkedIn → newsletter).
  6. Expert Techniques — Ogilvy headlines for subject lines, GaryVee Pillar→Micro for distribution, Brunson Epiphany Bridge for storytelling, Hormozi Value Equation and Suby HVCO for on-demand topics.
  7. Visual Generation — Pillow-based images matching AllAboutPMM visual style.

Architecture

creative-writer/
├── CLAUDE.md                          # Orchestrator v3.0 — routes to skills by mode
├── skills/
│   ├── generate/SKILL.md             # Multi-format content generation
│   ├── distribute/SKILL.md           # Atomize pillar → micro (newsletter → all)
│   ├── expand/SKILL.md               # Expand micro → pillar (LinkedIn → newsletter)
│   ├── showcase/SKILL.md             # Client demo / portfolio generator
│   ├── research/SKILL.md             # Deep research on a theme
│   ├── critique/SKILL.md             # Rubber Duck Escalator
│   ├── suggest-themes/SKILL.md       # Brain-informed theme suggestions
│   ├── visual/SKILL.md               # Visual brief + image generation
│   ├── ingest/SKILL.md               # Add new content to the system
│   └── scraper/SKILL.md              # Scrape Substack/Beehiiv newsletters
├── experts/                           # Tier 3: technique libraries (max 2 per request)
│   ├── ogilvy-headlines.md           # 9 headline principles, Big Idea test
│   ├── garyvee-distribution.md       # Pillar→Micro, JJJRH, platform-native rules
│   ├── brunson-storytelling.md       # Epiphany Bridge (7-step), Hook Story Offer
│   ├── hormozi-offers.md             # Value Equation, Grand Slam Offer, MAGIC naming
│   └── suby-leadgen.md              # HVCO, Godfather Offer, Larger Market Formula
├── references/                        # Protocol docs loaded by tier
│   ├── rubber-duck-escalator.md      # [Tier 1] 5-phase critique + format calibration
│   ├── generation-angles.md          # [Tier 1] 3 angles + hooks + format adaptations
│   ├── research-protocol.md          # [Tier 2] Research pipeline
│   ├── linkedin-practices.md         # [Tier 2] LinkedIn best practices
│   ├── twitter-practices.md          # [Tier 2] Twitter/X thread practices
│   ├── newsletter-practices.md       # [Tier 2] Newsletter article practices
│   ├── carousel-practices.md         # [Tier 2] LinkedIn carousel practices
│   └── format-adaptation-matrix.md   # [Tier 2] Cross-format adaptation rules
├── data/
│   ├── brain/pmm_brain.json          # PMM Brain v3.0
│   ├── output/                       # Generated content by format
│   │   ├── linkedin/
│   │   ├── twitter/
│   │   ├── newsletter/
│   │   └── carousel/
│   ├── scraped/                      # 13 newsletter JSON files (865 posts)
│   ├── user/
│   │   ├── my_posts.docx
│   │   └── visuals/                  # 20 reference images
│   └── writers/
├── scripts/
│   ├── fetch_missing_content.py      # Fetch full-text from Substack API
│   └── create_obsidian_posts.py      # Convert JSON → Obsidian markdown notes
├── src/                              # Python utilities
│   ├── schemas/                      # Pydantic data models
│   ├── brain/                        # brain_store.py, brain_injector.py
│   ├── storage/                      # SQLite + ChromaDB
│   ├── config/                       # Settings
│   └── ingestion/                    # .docx parsing, URL scraping
└── main.py                           # Utility CLI

How Claude Code Uses This

This is NOT a traditional Python app with API calls. Claude Code IS the LLM engine. The CLAUDE.md file acts as an orchestrator that routes requests to skills, loads the right references and expert techniques, and injects the right Brain sections into each task.

Content Formats

Format Target Key Feature
LinkedIn 800-1200 characters 3 angles, 5 hook types, visual for frameworks
Twitter/X 7-10 tweets, 280 chars each 4 thread types, character validation per tweet
Newsletter 1500-3000 words Deep pillar format, 3+ evidence anchors, subject lines
Carousel 5-10 slides 1080x1080, AllAboutPMM visual design, 1 point per slide

Usage (via Claude Code)

# Generation
"Write about {theme}"                    → LinkedIn (default, 3 angles)
"Write a newsletter about {theme}"       → Newsletter article
"Write a thread about {theme}"           → Twitter/X thread
"Write about {theme} for all formats"    → Newsletter first, then distribute to all
"Create a carousel about {theme}"        → Carousel brief + visual

# Distribution
"Distribute this"                        → Atomize newsletter → LinkedIn + Twitter + Carousel
"Expand this into a newsletter"          → Turn LinkedIn post → full newsletter article

# Research & Ideation
"Research {theme}"                       → Deep research → Brief + Gap Map
"What should I write about?"             → Brain-informed theme suggestions
"Critique this: {draft}"                → Run Rubber Duck Escalator

# System
"Add this content: {url}"               → Ingest into Brain
"Create client showcase"                 → Portfolio demo for prospects

PMM Brain (v3.0)

Dimension Count
Newsletter sources 13
Raw post notes (Obsidian) 861
Synthesized mental models 37
Evidence bank stats 69
Topic depth layers 11
Emerging beliefs tracked 26

Expert Technique Libraries

Extracted techniques (no voice mimicry) from 5 marketing experts:

Expert Techniques Use Case
David Ogilvy 9 headline principles, Big Idea test, Research-First Discipline Newsletter subject lines, carousel covers
Gary Vaynerchuk Pillar→Micro model, Context > Content, JJJRH ratio Distribution, multi-format generation
Russell Brunson Epiphany Bridge (7-step), Hook Story Offer, Attractive Character Story-led angles, newsletter expansion
Alex Hormozi Value Equation, Grand Slam Offer, MAGIC naming Posts about pricing, packaging, offers
Sabri Suby HVCO rules, Godfather Offer, Larger Market Formula Posts about lead gen, acquisition

Quality Gate: Rubber Duck Escalator

5-phase critique with format-adapted scoring:

Phase Lens What It Catches
MIRROR Experiential authenticity "Is this lived or Googled?"
PROBE Courage "What uncomfortable truth is being dodged?"
CHALLENGE Rigor "Where is the logic soft?"
ILLUMINATE Insight quality "Is the insight earned and non-obvious?"
CRYSTALLIZE Voice + AI decontamination "Does it sound like THIS person (not AI)?"
  • Threshold: All 5 phases must score 8+/10
  • Max 3 rewrite iterations before flagging to user
  • AI decontamination: 24 banned patterns (word, sentence, structure level)
  • Format calibration: Newsletter demands more evidence, Twitter checks character counts, Carousel checks visual consistency

Obsidian Vault

All research lives in an Obsidian vault at ~/Documents/Knowledge-brain-Obsidian/Knowledge Brain/PMM Brain/:

  • Raw Posts/ — 861 full-text newsletter notes with YAML frontmatter and wiki-links
  • Mental Models/ — 37 synthesis notes
  • Evidence Bank/ — 6 category notes
  • Topic Layers/ — 11 depth notes
  • Canvas Maps/ — Visual theme clusters, contrarian views, newsletter synergy

Connected: skills-test Project

The PMM Brain connects to /Users/sourav/skill-test/ which has:

  • 5 expert marketing personas (Ogilvy, GaryVee, Brunson, Hormozi, Suby) with full voice + frameworks
  • AI Super Team advisory board mode
  • PMM Brain Showcase skill for client demos and freelance outreach
  • 41 goose-skills capabilities

Utility CLI

python main.py init-db          # Initialize SQLite + ChromaDB
python main.py parse-docs       # Parse .docx files into structured data
python main.py scrape-urls      # Scrape URLs from data/writers/urls.txt
python main.py show-brain       # Display current Brain stats

About

PMM Content Distribution Engine — Newsletter scraper, Obsidian knowledge brain, multi-format content generation powered by Claude Code

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages