System Architecture

Under the hood of the
AI Toolkit.

We don't rely on generic "magic." Our systems are built on proven prompt engineering structures and optimized for specific Large Language Models (LLMs).

Explore Core Models
Context Data Input
Workflow Logic Processing
Optimized Result Output

Optimized for Top-Tier LLMs

Our workflows are not generic. We include specific prompt variations tuned to the strengths of the three major AI models.

ChatGPT
Model: GPT-4o
Native
  • Best For: Logic & Reasoning
  • Format: Markdown

The toolkit uses structured markdown tables and "Chain of Thought" prompting to leverage GPT-4o's superior logical reasoning.

Optimization: Markdown Tables & Step-Logic
Claude
Model: 3.5 Sonnet
Native
  • Best For: Writing & Nuance
  • Format: XML Tags

Claude excels at tone. We provide XML-tagged prompt variations that help Claude understand specific brand voice guidelines.

Optimization: XML <tag> Structure
Gemini
Model: 1.5 Pro
Compatible
  • Best For: Large Context
  • Format: Direct Instruct

Great for analyzing huge sets of data. Our Ops System includes instructions specifically for feeding Gemini large CSV files.

Optimization: Large Context Windows
Note: All systems are also backward-compatible with free models (GPT-4o Mini, Claude Haiku), though results may vary in depth.
Framework Logic

Engineering, not guessing.

A simple prompt gets a simple answer. Our systems use advanced prompt engineering techniques to force the AI into specific behaviors, ensuring consistent, high-quality output every time.

🔗

Chain-of-Thought (CoT)

We force the AI to "think" step-by-step before answering, reducing hallucinations and logical errors.

🎯

Few-Shot Prompting

We provide the AI with perfect examples of what "good" looks like, so it mimics your desired style immediately.

🛡️

Negative Constraints

We explicitly tell the AI what NOT to do (e.g., "Do not use hashtags," "Do not be robotic"), saving you editing time.

1# Role & Context Definition
2Role: "Senior Copywriter"
3Task: "Draft Email Sequence"
4# Dynamic Variables
5Context = {User_Business_Data}
6Tone = {Brand_Voice_Guide}
7# Chain of Thought Logic
8Step_1: Analyze target audience pain points.
9Step_2: Draft subject lines based on curiosity.
10Step_3: Critique draft against constraints.
11# Negative Constraints
12STOP: Do not use emojis. Do not sound salesy.
13OUTPUT: Return only the JSON object.
1. Business Profile (Input)
{Brand_Name} May Agency
{Target_Audience} E-com Owners
{Tone_Voice} Direct, Bold
INJECTED
2. Tailored Output

"Welcome to May Agency. We help E-com Owners scale faster. Stop wasting time..."

Context Architecture

Generic inputs give generic results.

Most people fail with AI because they don't give it context. Our systems include a dedicated "Context Vault" that injects your specific business DNA into every single workflow automatically.

  • ✓ Brand Voice Calibration

    We teach the model your specific writing style, banned words, and formatting preferences.

  • ✓ Product Knowledge Base

    Feed the system your pricing, features, and unique selling points so it never hallucinates your offer.

  • ✓ Audience Avatar Profile

    Define exactly who you are talking to (fears, desires, pain points) for hyper-relevant copy.

1
Initial Draft Raw

The AI generates a standard response based on your context variables.

2
Critique Agent Refining

A second logic layer reviews the draft against your negative constraints and tone guide.

3
Final Polish Ready

The approved content is reformatted and presented as the final output.

Quality Control

The "Self-Correction" Workflow.

Average prompts accept the first answer the AI gives. Our system doesn't. We use a recursive logic structure that forces the AI to critique its own work before showing it to you.

"It's like having a senior editor review every piece of copy before it hits your desk. The result is content that needs 90% less editing."
Model Agnostic

Built for GPT-4.
Ready for GPT-5.

The AI landscape changes fast. Our toolkit separates the "Business Logic" from the "Generation Model." When a new, smarter model is released, your systems don't break—they just get faster.

  • ⚡

    The Logic Stays

    Your workflows, SOPs, and context variables remain the same. Only the underlying engine changes.

  • 🔄

    Free Updates

    We actively monitor model releases. If a prompt needs tweaking for a new model, we update the master file instantly.

May Agency Toolkit

Business Logic Layer

GPT-3.5
GPT-4o
NOW
GPT-5

The Asset Library

No proprietary software. No monthly fees. Just raw, universal file formats you can import into the tools you already use.

.json

Prompt Schemas

Raw JSON structures for the advanced prompt chains. Import directly into AI studio interfaces or copy-paste the logic blocks.

Size: 1.2 MB Universal
.notion

Workspace Templates

Pre-built dashboards for Content, Support, and Ops. Includes One-Click "Duplicate to Workspace" links.

Blocks: 50+ Interactive
.csv

Tracking Sheets

Spreadsheet templates for campaign tracking and metrics. Compatible with Excel, Google Sheets, and Airtable.

Rows: Unlimited Structured
.pdf

SOP Documents

Standard Operating Procedures in PDF format. Clean, printable manuals to hand off to your team or VAs.

Pages: 100+ Print Ready
.drawio

Logic Maps

Visual flowchart files showing the decision trees of every system. Editable in draw.io or any diagram tool.

Editable Visual
.md

Install Guide

A "Readme" style setup guide in Markdown. Instructions on how to set up your variables and start the systems.

Step-by-Step Text

Why "Basic Prompting" Fails

The difference between a hobbyist using ChatGPT and a business using it is structure.

Standard User

  • ✕ Generic, vague context
  • ✕ Unstructured plain text output
  • ✕ Frequent hallucinations
  • ✕ Inconsistent tone & voice
  • ✕ One-off, single use
VS

May System

  • ✓ Injected Business Profile
  • ✓ Tables, Code, & Markdown
  • ✓ Self-Correcting Logic
  • ✓ Calibrated Brand Voice
  • ✓ Reusable Asset Library
Technical Specs

Implementation Details

Specific answers regarding security, API usage, and enterprise compatibility.

Docs Status: Up to Date
Models: GPT-4o / Claude 3.5
No. The toolkit is designed to work directly inside the standard ChatGPT (Plus) or Claude.ai interfaces. You do not need a developer account, API keys, or Python knowledge.
Yes. Since you run these prompts in your own private account, we (May Agency) never see your data. For maximum privacy, we recommend turning off "Chat History & Training" in your ChatGPT settings.
Yes. The prompts are text-based. You can copy/paste them into the ChatGPT or Claude mobile apps. However, for the Setup Phase (editing variables), we strongly recommend using a desktop computer.
Absolutely. You receive the raw source files. You are encouraged to tweak the "Brand Voice" and "Context" sections to perfectly match your evolving business needs.

Ready to Deploy?

Get immediate access to the complete asset library, source files, and installation guides. Start building your AI infrastructure today.

Access Documentation Review Specs