Voice Control: Voice-Driven AI Development Workflows

Training materials for Austin Public Health's Voice-Driven AI Development Workflows series.

---

Session 1 Presentation (January 2025)

Voice_AI_Workflows_Part1_Final.pptx — Main presentation deck for today's class

Supplemental Decks

Deck	Description
Claude Code Setup Guide	12-slide guide to Claude Code configuration
Cursor Azure Setup Guide	10-slide guide to Cursor with Azure OpenAI
VS Code AI Ecosystem Guide	12-slide overview of VS Code AI extensions
Voice-First Project Workflow	13-slide bonus: 10-phase project structure

---

Quick Start

cd redesign
npm install
npm run build:workflow    # Build the 13-slide workflow deck
npm run build:core        # Build the 36-slide core presentation
npm run build:full        # Build the 82-slide extended deck

Generated PowerPoint files appear in redesign/dist/.

---

Presentation Decks

Deck	Slides	Purpose	Build Command
Core Deck	36	Session 1 main presentation	npm run build:core
Full Deck	82	Extended reference version	npm run build:full
Workflow Deck	13	Bonus: 10-phase project structure	npm run build:workflow

---

About This Project

This repository was built almost entirely through voice-driven development—the same workflow it teaches. Every slide, script, and documentation file was created by speaking to an AI coding agent (Claude Code) using SuperWhisper for voice input.

How It Was Built

1. Voice brainstorming — Speaking ideas, getting pushback, refining concepts
2. Iterative slide creation — Describing layouts verbally, reviewing renders, adjusting
3. Design system emergence — Patterns crystallized through conversation, then codified
4. Continuous commits — Every conversation checkpoint saved to git
5. Verbal code review — "This slide is too cramped, give me more breathing room"

---

Workshop Series Roadmap

This is a 3-part training series teaching voice-driven development workflows for government technologists.

Session 1: Foundation (Current)

Voice Tools & SuperWhisper Configuration — 90 minutes

• Voice tool landscape (SuperWhisper vs. Wispr Flow)
• Security and PHI considerations for government use
• SuperWhisper deep dive: local models, custom modes, vocabulary
• Workshop: Building 5 custom modes for your workflow

Session 2: Application (February)

Voice-Driven Agent Development

• End-to-end project workflow using voice
• Working with AI coding agents (Claude Code, Cursor, etc.)
• Live build: Simple content aggregator application

Session 3: Advanced Patterns (March)

Agent-to-Agent Workflows

• Multi-agent architectures
• Task decomposition and delegation patterns
• Long-running autonomous workflows

---

Repository Structure

voice-control/
├── README.md                 # This file
├── CHANGELOG.md              # Version history
├── LICENSE                   # MIT License
├── redesign/                 # Build system and source files
│   ├── src/
│   │   ├── slides/           # Session 1 slides (82 HTML files)
│   │   ├── slides-workflow/  # Bonus workflow deck (13 HTML files)
│   │   └── supplements/      # Workshop exercise handouts
│   ├── lib/                  # HTML to PowerPoint conversion library
│   ├── scripts/              # Build scripts
│   ├── design-system/        # Design tokens and theme files
│   ├── docs/                 # Planning and research docs
│   ├── dist/                 # Generated PowerPoint files
│   └── package.json          # Node.js dependencies
├── unpacked/                 # Reference: unpacked PPTX structure
└── *.pptx                    # Legacy/reference presentations

---

Workshop Supplements

Hands-on exercises available in redesign/src/supplements/:

1. Model Cheat Sheet — Voice model comparison (local vs. cloud)
2. Mode Cheat Sheet — SuperWhisper mode quick reference
3. Local Email Mode — PHI-safe email dictation exercise
4. Cloud Email Mode — Wispr Flow formatting exercise
5. Prompt Enrichment — Structuring prompts for AI tools
6. Teams Chat Mode — Workplace messaging optimization
7. Custom Vocabulary — Recognition tuning for your domain
8. Agentic Handoff — Structured task specs for AI agents

---

Design System

The slide design uses a consistent visual language:

• Dimensions: 960×540px (16:9 aspect ratio)
• Primary color: #1a1f2e (dark navy)
• Accent color: #c9a227 (gold)
• Typography: Georgia serif for headings, system sans for body

---

Prerequisites

• Node.js 18+ — Required for slide conversion
• Playwright — Installed automatically via npm

---

License

MIT License — See LICENSE for details.

---

Contributing

This is training material for Austin Public Health. For updates, corrections, or questions about the voice-driven development workflow, contact the AI Technology Lead.

---

Built with voice. Shipped with confidence.