jcohen66

Model size must fit your memory

There are only two numbers you need to know about your machine.

Windows/Linux GPU VRAM

The memory in your graphics card.

• RTX 4090 has 24GB
• RTX 4090 has 32GB
• This is a hard ceiling.
• More tps

jcohen66

https://www.youtube.com/watch?v=7zZy1QTvokM

CLAUDE.md - Internal-OS

Knowledge management and agent orchestration for personal brant content creation.

How This Repo Works

• .claude/skills/ -- Slash-command workflows (scripts, newsletters, thumbnails, etc.)
• .claude/agents/ -- Consultant personas loaded by skills and agent tool
• knowledge/ - Three-layer knowledge base (see rules below)
• projects/ - Active and completed work (youtube, shorts, newsletters, linkedin)

jcohen66

The Spec

It is how you deliver your understanding to AI in a format it can use. A term you may have heard is "Claude's Plan Mode" to create a plan before building anything.

Uncover The Goal

Is the conclusion you are trying to draw. The decisions the report drives.

Interview me to identify the goal of this project.

Be Agile With How You Work

You break that same task into small buckets and show the result throughout the process.

jcohen66

Gem: Product Messaging Gem

NotebookLM can't cite specific files that it grounds from.

Identity

You are a Product Messaging Engine. Your job is to transform verified truth into structured, persuasive communication outputs while preserving factual integrity and scope boundaries.

Rules

Source Rule

jcohen66

Create an illustrated image comparing a messy uncut lawn next to a perfect looking grass lawn on the right.

Ask me 3 to 5 clarifying questions so you are 99% certain of what I want you to do.

jcohen66

https://www.youtube.com/watch?v=XOCVdPIhAHg

Prompt

create an LLM knowledge base for "How to set up my mac to use claude code to use local Ollama instead of cloud models". Set up three folders: raw/ for source material I drop in, wiki/ whwere you organize and cross-reference everything with wikilinks, and a CLAUDE.md that acts as the schema, telling you how to structure pages, what conventions to follow, and how to process new sources. Incloude a learnings.md file that tracks what works and what doesnt after each session. Set up a hook so that every time a file is written to raw/, you automatically process it into wiki pages, extract key concepts, and cross-reference everything with wikilinks. Then take this file I'm

jcohen66

https://www.youtube.com/watch?v=XOCVdPIhAHg

Prompt

Create a CLAUDE.md for this project. Include what this project is, the folder structure, what I'm currently building, and common mistakes to avoid. Keep it under 50 lines.

jcohen66

https://www.youtube.com/watch?v=AfKoqFwC7Ew

Prompt

Based on what you learned about me, recommend 2 experts who'd make a phenomenal advisory board for my specific situation. Bias toward YouTube creators. For each: why they fit + their best 5 pieces of content I should ingest.

jcohen66

https://www.youtube.com/watch?v=P-Fmo_CCIbY

Starter

• 7B-8B models
• RTX 4060 Ti with 16g of counter space.
• Not the 8g version
  • Load a real model, start a conversation, dishes pile up instantly
• Ryzen 5 cpu
• 64g RAM
• 2T SSD

jcohen66

Paradox

Traffic Controller vs Dispatcher

• Traffic Controller Mindset

  • Traditional software engineering is Deterministic. You own the roads, signals, and laws.

• Dispatcher Mindset

  • Agent engineering is Probablistic You give instructions to a driver tracking ambiguity.

• Trap