AI — 30-Day Mastery Mind Map

Overview

Comprehensive mind map by @MindBranches — “How to Master AI in 30 Days (The Exact Roadmap)”. Covers model selection, prompt engineering, context engineering, image/video generation, coding with AI, automation infrastructure, high-value workflows, open source models, custom knowledge systems, and personal AI assistants.

Foundational Mental Models — Understanding AI Processing

Context window: weights context to interpret words; give clear context for better results
Tokenisation: ~1 token = 3.5 chars or 0.75 words — affects costs and limits
Temperature: parameter (0-1) — set low (0) for factual/analytical, high (1) for creative
Hallucination: structural, not a bug — verify claims, use low temperature for facts, implement RAG systems

Model Selection Framework (January 2026)

Model	Best For
Claude (Anthropic)	Coding (Claude Opus 4.5+ reads codebase), Marketing/long-form writing, Spreadsheet/business analysis, Excel integration
Gemini 3 Pro (Google)	Research with 1M token context window, Native Google Search integration, Tasks requiring recent data or creative document analysis
Grok	Real-time X social analysis
Nano Banana Pro	Image generation
VEO 3.1	Video generation
GPT-5	General AI-sounding output

Format by Model:

Claude/Gemini: use JSON for structured data
Markdown: works well overall
Chain-of-Thought: “Let’s think through this step by step” for complex problems

Prompt Engineering 2026

System Prompt Formula:

Role — who the AI should be
Behaviour — how it should interact
Constraints — what to avoid
Output structure — format specifications

Four Strategies:

Write: save context in knowledge/code/reference files
Select: use RAG and dynamic retrieval vs dumping everything
Compress: summarise verbatim into before including
Isolate: separate threads/sub-agents for different contexts

Context Engineering (The 2025-2026 Skill)

Claude Projects in Practice:

Create persistent workspaces with uploaded documents accessible across conversations
One focused project per task beats one-size-fits-all
Upload files, write custom instructions, reuse hundreds of times

RAG for Non-Technical Users:

NotebookLM (Google) — free, no-code RAG
Upload PDFs, docs, YouTube videos for AI expert on specific content with citations
Grounds responses in actual data vs training data

Image Generation

Nano Banana Pro:

Perfect text rendering in images
Reasoning before rendering for intentional compositions
Self-grounding for factually accurate infographics
Natural language prompts: describe like briefing a photographer
Include subject, details, action, environment, composition, lighting, text requirements

Alternatives:

Midjourney V7 — most artistic/cinematic for stylised work
Flux — open-source for local generation

Video Generation

VEO 3.1 (Google):

Up to 60 seconds, 4K output, vertical format support
Use for finished clips with audio

Kling 2.6: Best cinematic realism

Best Practices:

5-10 seconds is reliable range
Budget 8-10 attempts per viable clip
Prompt like a director describing camera view, not narrative storytelling
Sweet spot: social shorts under 15 seconds, B-roll, product reviews

Coding with AI

For Developers:

Claude Code — runs in terminal, reads codebase, multi-file edits, creates commits
Cursor — AI-first IDE built on VS Code

For Non-Developers:

Lovable — natural language to complete web application without code
Bolt.new — rapid prototyping from plain English
Replit — browser-based development with AI assistance
“Vibe coding” — describe what you want, iterate based on results

n8n Platform (Automation & Infrastructure)

Open source, self-hostable, unlimited free workflows
Claude Code generates n8n configurations from natural language
Describe workflow → Claude Code generates config → deploy
Universal adapter for AI systems to connect external tools/APIs
Pre-built MCP servers for common services
n8n creates custom MCP servers from workflows

MCP (Model Context Protocol)

Universal adapter for AI systems to connect external tools/APIs
Pre-built MCP servers for common services
Create custom MCP servers from workflows

High-Value Workflows

Content repurposing — one post becomes multiple platform-specific versions
Customer feedback routing — sentiment analysis → automated ticketing system

Open Source Models (Current Leaders)

Kimi K2.5 (MoonshotAI) — 1 trillion parameters, 1/10th GPT cost
DeepSeek V3.2 — matches GPT-5, 90% lower training costs, self-hostable
GLM 4 (ZhipuAI) — strong coding
Mistral — fast, 1M token context at fraction of Claude price

Timeline:

6-12 months: consumer hardware runs capable local models
12-24 months: open source likely matches/exceeds closed models

Custom Knowledge Systems

Claude Projects Alternative:

Upload documents, reference automatically in conversations
More flexible for creating outputs vs just querying

Vector Database Approach:

Documents split into chunks → converted to embeddings → stored in vector database
Query becomes embedding → find similar chunks → LLM produces grounded answer

Personal AI Assistants

Clawbot:

Runs entirely on hardware (Mac/Windows/Linux or R2me VPS)
Connects to WhatsApp, Telegram, Slack, Discord, Signal, iMessage
Persistent memory across conversations
Can read/write files, control browsers, execute scripts, build extensions
Self-modifying: writes code to expand its own capabilities

The 30-Day Implementation Roadmap

Days 1-7 (Fundamentals):

Mental models, how AI processes information

Days 8-14 (Power & Context Engineering):

Multiply value of every interaction

Days 15-21 (Creative & Technical Tools):

Image/video/coding with immediate feedback

Days 22-30 (Advanced Integration):

Automation, RAG, personal knowledge systems

Single Highest-Leverage Move:

Build a Claude Project for a repetitive task
Upload relevant documents, write custom instructions
Run it 5+ hours weekly

Key Resources

Anthropic Prompt Guide (official docs)
OpenAI Tokenizer (visualise text to tokens)
Andrej Karpathy’s LLM videos
TypingMind/Typing AI
NotebookLM — free RAG system
OpenRouter — unified access to all major models
ClawBot (GitHub) — open-source personal assistant

Critical Insight

“The gap between AI-fluent and AI-confused widens monthly. The window is time-sensitive — compound over time. The window is time-aware: 30 days from now, students using AI autonomously are still collecting bookmarks. The roadmap exists, the choice is yours.”