AI Admin System
AI-Powered Admin & Worksheet Upload
Intelligent content management system reducing publishing time by 60%

What It Does


Why I Built This

For nearly four years at All Kids Network, I watched the content publishing workflow evolve organically—adding worksheets, categorizing them, uploading images, and tagging resources for educators worldwide. The process worked, but it was manual, repetitive, and time-consuming. Every worksheet required navigating multiple forms, double-checking category assignments, and ensuring metadata accuracy.

In March 2024, I started tracking exactly how long it took to publish a single worksheet: 10 minutes on average. That might not sound like much, but multiply it by dozens of worksheets per month, and it became clear we were losing hours to administrative overhead. I knew we could do better. I began documenting the bottlenecks, analyzing the workflow, and imagining what a modern, AI-assisted admin system could look like.

This wasn't just about saving time—it was about creating a system that could scale, that made categorization intuitive, and that gave content creators the tools they deserved. I wanted to build something that felt seamless, intelligent, and genuinely helpful.


How It Works

The system is built with React and TypeScript, providing a responsive, component-based architecture that adapts to different resource types. The admin interface uses a pane-based layout system, allowing users to work with multiple files simultaneously while maintaining clear context switching.

File uploads leverage drag-and-drop functionality with visual feedback and preview support. The categorization system uses tree-structured checkbox inputs with automatic parent-child relationship handling—selecting a child category automatically highlights parent categories, and partial selection states indicate mixed hierarchies. All form inputs follow a standardized initialization pattern for consistency across the system.

An integrated AI assistant helps with content categorization, metadata suggestions, and workflow guidance. The backend connects to AWS services and uses DynamoDB for resource storage with support for branching conversation histories. The entire system was designed to be mobile-responsive, ensuring content managers can work from any device.


Impact

By the numbers:

What changed:


Challenges & Solutions

The most complex challenge was designing the tree checkbox input system. Educational resources can belong to multiple overlapping categories—a Halloween-themed math worksheet might be tagged under "Holidays > Halloween," "Subjects > Math," and "Grades > 3rd Grade" simultaneously. I needed a UI that made these hierarchical, multi-select relationships intuitive without overwhelming users.

I solved this by implementing automatic parent selection logic: when you select a child category, the system automatically highlights the parent categories to show the complete path. Partial selection states (indicated by visual styling) show when only some children in a category are selected. The entire tree structure is alphabetically sorted and supports keyboard navigation for accessibility.

Another major challenge was mobile responsiveness. Content managers don't always work from desktops—they might be uploading resources from tablets or phones. I conducted extensive research into mobile-friendly form patterns and implemented collapsible sidebars, touch-optimized controls, and adaptive layouts that gracefully scaled from phone screens to wide desktop monitors.


What I Learned

This project taught me the importance of workflow analysis before writing a single line of code. By spending time in March documenting the existing process, benchmarking upload times, and identifying friction points, I built a system that addressed real pain points rather than imagined ones. The 60% time reduction wasn't magic—it came from systematic observation and prioritization.

I also learned the value of design consistency at scale. Early in the project, I created a style guide and naming convention for all components. That upfront investment paid dividends as the system grew more complex—every new feature followed established patterns, making the codebase predictable and maintainable.

Working with AI integration was eye-opening. I learned how to design systems that augment human decision-making without replacing it. The AI assistant suggests categories and metadata, but the final decisions remain with the content creator. That balance between automation and control is critical for building tools people actually trust and use.

Future improvements:


Links