Prompt Engineer / Prompt Generator

A documentation-driven Custom GPT system designed to craft and enhance AI prompts.

DEMO LINK • GITHUB LINK

Overview

• Role: Creator / Prompt Engineer
• Stack: Custom GPT, Diverse Prompt Formats, JSON
• Key Result: Successfully designed a code-free, highly extensible prompt generation engine.
• Summary: This project marked my very first time applying formal prompt engineering. It is a modular, documentation-driven system for generating, refining, and managing prompts, where all logic and workflow are managed through structured files containing tons of different formats, templates, and paradigms for clarity, maintainability, and extensibility.

The Problem

• Challenge: Users often struggle to get the best possible outputs from AI models due to poorly structured prompts. The challenge was creating a system to automate the optimization of these prompts without requiring traditional code.
• Audience: Anyone who uses AI and wants to significantly enhance the quality and accuracy of the AI's output.
• Constraints: Ensuring all logic, workflows, and techniques could be parsed effectively by a Custom GPT relying entirely on text-based configurations and extensive format blueprints rather than executable code.

Approach & Decisions

• Architecture: A declarative, multi-format architecture where all system instructions are stored in interlinked files.
• Key Decisions: Decided to map out workflows, context, and commands using a strict folder structure (context/, commands/, templates/) to keep the GPT's knowledge base organized and modular.
• Trade-offs: Exchanged the computational power of a traditional programmed web app for the immediate, conversational accessibility of a Custom GPT platform fueled by a massive library of formats.
• Data Flow: User provides a basic prompt -> GPT references structured templates, paradigms, and formats -> GPT outputs a highly optimized, professional prompt.

Key Results & Impact

• Outcome: A highly effective tool that consistently upgrades basic user instructions into professional, proven prompt formats.
• Impact: Provides significantly enhanced outputs from AIs and LLMs by ensuring the input perfectly matches the best structural format for the task.

Technical Deep Dive

• Notable Details: Curated a rich context of industry best practices, housing tons of formats and paradigms directly into the GPT's knowledge base.
• Verification: Relies on scientifically and empirically proven prompt engineering formats and techniques that have already been tested to enhance AI outputs.
• Optimizations: Separated domain-specific context from active commands (context.md vs commands.md) to optimize the GPT's context window usage.
• Code Quality: Maintained strict structural consistency across all format blueprints to ensure the GPT parses instructions predictably.
• Links: See the GitHub Repository for the file structure.

Reflections

• Differently: Instead of keeping it purely as a standalone GPT, I decided to take these concepts further by implementing an AI subagent in my oz: multiagent project to automate output enhancement directly within my own apps.
• Successes: Successfully proved that a robust, modular logic system can be built entirely through declarative markdown and prompt engineering techniques.
• Lessons: Structured formats are the absolute key to predictable and high-quality LLM outputs.

My Contributions

• System Design: Designed the modular, documentation-driven architecture for clarity and maintainability.
• Workflow Engineering: Defined declarative workflows, commands, templates, and profiles in markdown files.
• Best Practices: Curated context, techniques, and paradigms for prompt engineering.
• Documentation: Authored and maintained all project documentation and workflow files.

Project Architecture

This project uses a Documentation-driven, modular file structure. There is no code execution; all logic resides in markdown and config files.

• Component Organization:
  • main_instruction.md: Central JSON configuration and system instructions
  • context/context.md: Domain-specific context, best practices, formats, paradigms, techniques
  • commands/commands.md: All available commands, syntax, and workflow automation
  • profiles/: Agent and workflow profiles
  • templates/: Prompt templates, subcomponents, techniques, examples

Demo Preview

You can view the live Custom GPT via Prompt Engineer / Prompt Generator (Live Custom GPT). All features and workflows are available to explore interactively.

• Platform Views: Web (OpenAI ChatGPT Custom GPT)
• Functionality Shown:
  • Interactive prompt task, format, technique, and paradigm selection
  • Real-time workflow management and validation
  • Modular, documentation-driven architecture

Hosting Note:
The live Custom GPT is available on OpenAI. Source code and documentation are on GitHub; no code execution or build system is present.