Skip to content

mumima808/AI-Prototyping-Stack

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

AI Prototyping Stack

A pragmatic methodology for building AI workflows — from idea to production.

"If it works, it's good enough."


Overview

This repository documents a battle-tested approach to AI workflow design and prototyping, rooted in a simple but powerful philosophy: validate the idea first, optimize later.

The methodology was born from a real-world factory management system — a production traceability tool built entirely on WordPress's post structure — prototyped originally in Scratch. That same thinking now extends to AI-powered automation stacks.


The Stack

┌─────────────────────────────────────────────────────┐
│  LAYER 1 · THINK & PROTOTYPE                        │
│  Scratch / TurboWarp                                │
│  Validate logic and state transitions visually.     │
│  Constraints spark creativity.                      │
├─────────────────────────────────────────────────────┤
│  LAYER 2 · AI PROTOTYPING                           │
│  Dify                                               │
│  Prototype the AI layer without writing code.       │
│  Hand off to non-engineers. Iterate fast.           │
├─────────────────────────────────────────────────────┤
│  LAYER 3 · PRODUCTION AUTOMATION                    │
│  n8n                                                │
│  Connect APIs, webhooks, and databases.             │
│  Break limits with JavaScript/Python code nodes.    │
└─────────────────────────────────────────────────────┘

Tool Roles

Scratch / TurboWarp — Thinking Engine

The official Scratch environment enforces constraints that paradoxically free your thinking. Once the logic is validated, TurboWarp compiles the project to JavaScript for dramatically faster execution.

Scratch TurboWarp
Role Idea validation Performance testing
Speed Baseline Up to ~10–18x faster
Editor
Extensions Standard Extended

Note: Forkphorus is now in maintenance mode. TurboWarp is the recommended MOD.


Dify — AI Layer Prototyping

Dify sits at the intersection of no-code accessibility and LLM power. It's the right tool when you need to:

  • Prototype RAG pipelines and agent workflows quickly
  • Hand a working AI app to a non-engineer
  • Swap LLM providers without rewriting logic
  • Test prompt strategies with a visual flow editor

Think of Dify as the packaging layer — it wraps AI capabilities into something usable before committing to a full engineering implementation.


n8n — Production Workflow Engine

Where Dify is for AI prototyping, n8n is for connecting systems at scale. Engineers will feel at home because:

  • Data flows between nodes as visible, inspectable JSON
  • JavaScript/Python code nodes provide an escape hatch for complex logic
  • Self-hosting via Docker gives full infrastructure control
  • Webhooks and REST APIs are first-class citizens

Use Case: Video Production Pipeline (Vrew-Inspired)

Vrew by VoyagerX (Korea) represents the current frontier of AI video editing — it treats video as text, enabling transcript-based editing, auto-captions, AI narration, and one-prompt video generation.

Since Vrew does not currently expose a public API, the equivalent pipeline can be assembled from open components:

VREW CAPABILITY          →  OPEN STACK EQUIVALENT
─────────────────────────────────────────────────
Speech recognition       →  Whisper (OSS) / AssemblyAI
Auto-captioning          →  Whisper + custom formatter
AI narration             →  ElevenLabs / OpenAI TTS
Script generation        →  Claude / GPT via Dify
AI image generation      →  Stable Diffusion / DALL·E
Video cutting & joining  →  FFmpeg

Prototyping flow:

① Scratch / TurboWarp  →  Validate state machine for video pipeline
② Dify                 →  Prototype AI nodes (script, captions, voice)
③ n8n                  →  Wire APIs into a production-grade workflow

Origin: Factory Management System

The philosophy of this repository was proven in production before AI tools existed.

A factory traceability system was built by repurposing WordPress's post structure as a production data model:

WordPress Concept     →  Factory Management Meaning
──────────────────────────────────────────────────
Post ID               →  Lot / Serial number
Category / Tag        →  Process stage / Product type
Created / Updated at  →  Timestamp / Trace history
Custom fields         →  Arbitrary manufacturing data
Post status           →  Process state (draft → published → archived)

Prototype path: Scratch → WordPress → Running in production today.

This is the proof of concept for the entire methodology.


Roadmap

  • Dify × WordPress REST API integration (natural language queries over factory data)
  • Automated video captioning pipeline with Whisper + n8n
  • TurboWarp state machine template for workflow prototyping
  • Dify RAG pipeline for factory manual Q&A
  • n8n workflow: audio → transcript → structured post → WordPress

Reference Tools

Tool URL Purpose
Dify https://dify.ai LLM app development platform
n8n https://n8n.io Workflow automation
TurboWarp https://turbowarp.org High-performance Scratch MOD
Vrew https://vrew.ai AI video editing (reference implementation)
Whisper https://openai.com/research/whisper Open-source speech recognition
AssemblyAI https://www.assemblyai.com Speech recognition API
ElevenLabs https://elevenlabs.io AI voice synthesis
FFmpeg https://ffmpeg.org Video processing

Methodology at a Glance

Phase Tool Goal
Conceive Paper / mind Define what to build
Validate Scratch / TurboWarp Make it work, fast
AI prototype Dify Test the AI layer, no code
Productionize n8n Connect APIs, automate
Persist WordPress / GitHub Store with battle-tested tools

Philosophy

This stack deliberately avoids chasing every new framework. The through-line is always the same:

  1. Validate the logic in the simplest possible environment (Scratch).
  2. Prototype the AI layer without committing to code (Dify).
  3. Automate for production with full engineering control (n8n).
  4. Store with boring, proven technology (WordPress, GitHub).

New tools are evaluated as rungs on this ladder, not replacements for it.


📄 Japanese version: README_jp.md

Generated from a Claude conversation session — April 2026. Resume this project by sharing this README in a new session.

About

A pragmatic methodology for building AI workflows — from idea to production. "If it works, it's good enough." | 「動けばOK」哲学によるAIワークフロー設計メソッド。

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors