Slide Maker: An Intelligent Presentation Builder Based on Conversational AI

The open-source slide-maker project by CUNY AI Lab enables collaboration with AI via natural language dialogue to create professional presentations. Its core features include:

• Conversational creation workflow, allowing users to focus on content rather than formatting
• Real-time editing and multi-model support (Anthropic Claude, AWS Bedrock, etc.)
• Rich visualization components and interactive artifacts
• Open-source project with support for custom themes and deployment

This thread will introduce the project's background, technical architecture, functional features, and future outlook across different floors.

Project Background and Motivation

In modern work and academia, presentations are core information media, but traditional tools require significant time for typesetting and design. CUNY AI Lab developed slide-maker to reconstruct the creation process via conversational AI:

• Solve the tedious format adjustment issues of traditional tools
• Build a complete agentic workflow system that understands user intent and proactively provides suggestions
• Achieve a "what you ask is what you get" experience, representing a new direction for productivity tools

The project is not just simple template filling; it is a creation assistant deeply integrated with AI.

Core Architecture and Tech Stack

slide-maker uses a full-stack modern architecture:

• Frontend: SvelteKit + Svelte5 Runes, providing responsive UI and smooth interaction
• Backend: Hono framework (Node.js), with SQLite + Drizzle ORM for data persistence
• Authentication: Lucia v3, supporting email registration and *.cuny.edu domain restrictions
• Code Management: Turborepo monorepo, including three packages: web application, API service, and shared type definitions

Modular design enhances reusability and collaboration efficiency.

Conversational Creation and Visual Editing

Conversational Creation

• Describe requirements via natural language interaction; the system parses intent in real time to generate slides
• SSE streaming transmission, allowing users to instantly see AI's thinking process and results
• Support for multi-model switching (Claude, Bedrock, OpenRouter open-source models)

Visual Editing

• Three-panel layout: chat/outline, canvas editing, resource management
• Multiple layout templates (title page, two-column, grid cards, etc.)
• Drag-and-drop to adjust module order/size, TipTap rich text editing, undo/redo functions

Seamless switching between dialogue generation and manual refinement.

Content Components and Resource Integration

Content Components

• Diverse modules: heading, text, card, image, carousel, etc.
• 14 interactive artifacts: A* pathfinding, flocking simulation, Lorenz attractor, etc., with configurable parameters and adaptive sizing

Resources and Export

• Resource integration: Tavily/Brave search, Pexels image library, PDF/DOCX upload (AI parses content)
• Export: Self-contained ZIP file, supporting browser opening, keyboard navigation, and accessibility

Meets the visualization needs of teaching and technical presentations.

Theme System and Deployment Operations

Theme System

• 9 built-in themes (Studio Dark/Light, CUNY series, etc.), based on CSS custom properties
• Support for custom/derived themes, complying with WCAG AA/AAA contrast standards

Deployment and Operations

• Multi-environment configuration: development, pre-release (tools.cuny.qzz.io/slide-maker), production
• Architecture: Cloudflare + Caddy + Nginx + PM2
• Management features: usage statistics, token quota (1 million/month by default), user role management, login/registration rate limiting

Ensures secure and convenient deployment.

Summary and Outlook

slide-maker represents a new paradigm for AI-assisted creation tools:

• Deeply integrate AI into the entire creation process (intent understanding → real-time rendering → fine editing)
• Lower the threshold for presentation creation while maintaining professional output
• Value: An efficient tool for educators, technical speakers, and content creators

In the future, with the development of multimodal AI, such agentic tools will play a role in more creation scenarios.