Show of Hands AI Tools

Transforming Retail Operations with Specialized AI

Case Study

At Show of Hands Denver, adding a single product to the online store used to take 30 minutes of careful work. Employees had to find products on wholesale sites, copy descriptions written for store owners (not customers), rewrite everything to sound appealing, craft SEO metadata, generate searchable tags, and convert it all to HTML. All of that before adding the product to Artisan POS, which then uploaded it to Shopify. It was tedious, time-consuming, and required both creative writing skills and technical knowledge.

I directed AI agents to build four specialized bots that reduced this entire process to ~30 seconds—a 98% time reduction with consistently professional results. For the past several months, I've been using these tools myself to maintain and grow the online store, transforming it from a sparse catalog to a fully stocked, ever-expanding inventory. While they haven't yet been rolled out to the broader team, my goal is to empower employees with these AI systems in the near future so they can focus more on customer service, inventory, and running the business—rather than tedious product entry.

This system is built on my Custom GPT System platform and represents one of the most successful real-world business applications of my AI work. These tools are used weekly, processing hundreds of products, and have fundamentally transformed how Show of Hands manages their online presence.

Overview

Goals

• Reduce product processing time from 30 minutes to under 1 minute per item.
• Eliminate the need for creative writing skills in product data entry.
• Generate consistent, professional, SEO-optimized product content automatically.
• Convert wholesale-focused descriptions into customer-friendly retail copy.
• Produce clean, traditional HTML formatting without manual coding.
• Create searchable metadata and tags that improve product discoverability.
• Make the system simple enough for any employee to use without technical training.

What I Learned

Show of Hands AI Tools taught me that even limited technical knowledge can have enormous impact in the right environment. When I started working at Show of Hands, I was just helping with product uploads — it wasn't my intention to build a generative AI system. But as soon as I saw how much time was required to rewrite descriptions, format HTML, and do SEO by hand, I realized how valuable even my beginner-level understanding of AI architecture could be.

The experience taught me that my skills don’t have to be “expert-level” to be useful. In the right setting, even a modest understanding of AI systems can completely transform a workflow. I learned not to hold back what I know just because I’m still learning.

The second big lesson was about constraint-based prompt engineering. I discovered that telling an AI what not to do is often more effective than describing what it should do. For example, banning markdown formatting, forcing raw HTML output, and setting strict rules for tone and length — all of that came from being aggressively explicit about what to avoid. That’s how you move from “creative” output to production-grade output that people can trust and paste directly into the site.

And finally, I learned the value of restraint. For six months, the system ran flawlessly on GPT-4o Mini — fast, efficient, and cheap. When GPT-5 came out, I switched everything over assuming it would be an automatic upgrade. It wasn’t. It ran slower, cost more, and required me to re-tune prompts that had already been dialed in. That was a good reality check: newer isn’t always better. If a tool is stable, affordable, and doing exactly what it’s supposed to, leave it alone. Upgrade when there’s a reason, not just because you can.

More than anything, this project showed me that building great systems isn’t about knowing everything — it’s about knowing what matters, applying it where it counts, and trusting that small, well-aimed skills can create massive change.

The Problem: A 30-Minute Workflow

Before the AI system, every product required employees to:

1. Find Products (2 min): Browse wholesale sites like Faire.com to discover inventory.
2. Copy Descriptions (1 min): Extract product information written for wholesale buyers, not retail customers.
3. Rewrite for Customers (10 min): Transform B2B language into engaging B2C copy — "Your customers will love..." becomes actual customer-focused descriptions.
4. Write SEO Descriptions (5 min): Craft compelling 150-character summaries with keyword integration.
5. Generate Tags (7 min): Create 12-16 searchable keywords for product discovery.
6. Format HTML (5 min): Convert everything to proper HTML with paragraphs, bold text, and bullet lists.

Total time: 30 minutes per product — if the employee was creative, focused, and experienced with web formatting.

The Solution: Four Specialized AI Bots

I directed AI to build four specialized bots, each bot was designed to handle one specific step in the product workflow. Together, they transform the 30-minute process into a 30-second pipeline.

The New Workflow: 30 Seconds

With the AI system, the process is dramatically simplified:

1. Copy to Dee (5 sec): Paste wholesale description into Dee → receive customer-friendly version.
2. Copy to Tedium (5 sec): Paste Dee's output into Tedium → receive formatted HTML.
3. Copy to Amelia (5 sec): Paste original description into Amelia → receive SEO description.
4. Copy to Gus (5 sec): Paste product details into Gus → receive searchable tags.
5. Paste to Website (10 sec): Copy all outputs directly into the product database.

Total time: 30 seconds per product — a 98.3% time reduction with consistently professional results.

Prompt Engineering: The Technical Foundation

The effectiveness of this system lies entirely in precision prompt engineering. Each bot's behavior is controlled through carefully crafted system contexts that ensure repeatable, reliable results.

Strategic Token Economy

Every instruction is optimized for clarity and brevity. The prompts achieve consistent results with minimal token usage by focusing on specific constraints rather than lengthy examples. This token efficiency translates directly to faster response times and lower API costs.

Context Architecture

Each bot uses a dual-layer system:

• systemContext: Defines role, behavior rules, and output constraints
• assistantContext: Provides additional guidance and response patterns

This separation allows precise control over bot personality (system) versus response style (assistant), enabling complex behaviors through simple configuration.

Constraint Engineering

The system's reliability comes from explicit exclusion rules that prevent unwanted outputs:

Temperature Precision

Temperature settings are strategically chosen for each bot's purpose:

• 0.1 (Dee, Tedium, Gus): Minimal variance ensures identical inputs produce identical outputs which is critical for consistency
• 0.7 (Amelia): Balanced creativity prevents repetitive SEO descriptions while maintaining quality

Repeatability Engineering

The prompts are designed so that identical input produces identical output every time. This consistency is what enables true automation — employees trust that the bots will always deliver professional results, eliminating the need for manual review.

Performance Metrics

These precision prompts directly enable the 98% time reduction. By constraining outputs to exact specifications, the system eliminates revision cycles, manual formatting, and quality inconsistencies. The prompts transform generative AI from a creative tool into a reliable production system.

Real-World Impact

The Show of Hands AI Tools are used daily and have fundamentally transformed store operations:

• 98% Time Reduction: Product processing dropped from 30 minutes to 30 seconds per item.
• Hundreds of Products Processed: The system has handled the entire online catalog expansion.
• Zero Technical Training Required: Any employee can use the bots without coding or SEO knowledge.
• Consistent Professional Quality: Every product receives the same level of polish and optimization.
• Eliminated Bottlenecks: Product uploads are no longer a time-consuming constraint on business growth.
• Cost Savings: Dramatic reduction in labor hours translates directly to operational savings.

These aren't theoretical improvements — this system is mission-critical infrastructure used every single day.

Technical Implementation

Built on Custom GPT System

The bots run on my Custom GPT System platform, leveraging its modular architecture for easy updates and maintenance. Each bot is defined by a simple JavaScript configuration file — no database, no complex deployment pipeline.

Technologies Used

• OpenAI GPT-4o: Powers all natural language processing and content generation
• Vanilla JavaScript: Lightweight, fast, no framework dependencies
• PHP: Server-side bot validation and routing
• Custom Prompt Engineering: Precision-crafted system contexts for each bot
• OpenAI TTS: Voice synthesis for audio feedback (optional feature)

Architecture Highlights

• Modular Bot Configs: Individual JavaScript files in /configs/ directory
• Unified Engine: Single bot-engine.js handles all API calls and state management
• Dynamic Loading: Bots load on-demand with cache busting for instant updates
• State Management: Conversation history, temperature settings, and feature flags per bot

Challenges and Solutions

Challenge: Preventing Markdown Formatting

Early versions of Dee would add bold and italic markdown formatting, which broke the HTML conversion process. The solution was adding explicit "CRITICAL FORMATTING RULE" instructions at the top of the system context with multiple redundant constraints.

Challenge: HTML Wrapper Pollution

Tedium initially wrapped output in full HTML documents (<html>, <body> tags), making copy/paste impossible. The fix was specific exclusion rules: "NO <html> or <body> wrappers" combined with "Raw HTML output only" instructions.

Challenge: SEO Description Repetition

Amelia was generating identical structures for similar products. Increasing temperature to 0.7 and adding style variation instructions ("vary your approach to ensure uniqueness") solved the repetition while maintaining quality.

Challenge: Tag Consistency

Gus would sometimes generate 8 tags, sometimes 20. Adding explicit numeric constraints ("12-16 SEO-friendly tags") and focusing on "one-word keywords" created predictable, usable output.

Other Things I Learned

• Temperature is Critical: The difference between 0.1 and 0.7 determines whether you get consistency or creativity — choose strategically.
• Redundancy Improves Reliability: Stating the same constraint multiple ways ensures the AI follows critical rules.
• Real-World Testing is Essential: Prompts that work in testing can fail in production — iterate based on actual usage patterns.
• Token Economy Matters: Concise, precise prompts respond faster and cost less — every word should serve a purpose.
• Modular Architecture Enables Rapid Iteration: Being able to update a single bot without touching others accelerated refinement.

Future Development Roadmap

Unified Pipeline

Currently, users interact with four separate bots sequentially. The next version will combine all four into a single API call that returns all outputs simultaneously — paste one product description, receive customer copy, HTML, SEO description, and tags in one response.

Batch Processing

Upload a CSV file with 50 products → receive a structured data file with all processed outputs in seconds. This will enable bulk inventory uploads without manual intervention, further reducing the time investment to near-zero.

Workflow Evolution

The system will evolve from "4 sequential bot interactions" (30 seconds) to "single-command complete product processing" (5 seconds). The ultimate goal: paste raw wholesale data, click one button, get everything formatted and ready for the website.

Why This Matters

This project demonstrates that AI isn't just about chatbots or creative experiments — it's about solving real business problems with measurable results.

The Show of Hands AI Tools prove that well-engineered prompts can transform 30-minute manual workflows into 30-second automated processes. The system has processed hundreds of products, eliminated operational bottlenecks, and enabled business growth without adding labor costs.

This is precision prompt engineering in production — reliable, repeatable, and genuinely transformative.