Use Cases and AI Architecture

Identifying Your First Use Case

This guide will help you identify potential AI applications for your specific needs and select the most appropriate AI architecture in Navigator.

Identifying Potential Use Cases

Before implementing an AI solution, consider what problems in your organization could benefit from automation or intelligence. Here are some common scenarios where AI can provide significant value:

  • Document Intelligence: If you have years of documentation that takes hours to search through, an AI expert trained on these documents could provide answers in seconds, saving significant time.
  • Compliance Monitoring: As a safety compliance officer, you could use AI to monitor security camera feeds and receive alerts when team members aren't wearing proper protective equipment in designated areas.
  • Automated Classification: For organizations with large image databases that need tagging by features (type, color, material, etc.), AI can automate this process, saving hours or days of manual processing.

Planning Your AI Solution

Once you've identified a use case, break it down into incremental steps:

  • Define specific goals - What exact problem are you solving?

  • Start broad, then refine - Begin with general categories before moving to specific details

  • Implement, test, and iterate - Build your solution progressively

Example: Art Supply Image Classification

Let's say you need to tag a dataset of art supply images:

Identify classification categories:

  • Item Color: Red, Green, Blue, Black, White...
  • Item Use: Drawing Supply, Painting Supply, Sculpting Supply...
  • Item Type: Paint Brush, Pencil, Paint, Canvas, Clay...

Implementation Steps:

  1. Start with broad classification:

    • Begin by training a model to identify Item Use (Drawing, Painting, or Sculpting Supply)

    • Test accuracy with a separate test set

    • Adjust or retrain until accuracy meets your needs

  2. Add more specific classification:

    • Train additional models for more specific tags (e.g., Item Type)

    • Test each model individually

    • Connect models in sequence and verify combined performance

    • Continue adding classifiers until all needed tags are implemented

Choosing the Right AI Architecture

Based on your use case and available data, select the appropriate AI architecture:

Large Language Model (LLM) Solutions

  • Choose when: Your data is primarily text-based (PDFs, Word documents, text files)
  • What it does: Creates an AI subject matter expert you can chat with through text prompts
  • Key benefit: Provides conversational access to knowledge in your documents

Computer Vision Solutions

Computer vision solutions require image data (still images or video) and come in two main types:

Image Classifiers

  • Choose when: You need to categorize entire images
  • What it does: Labels a complete image based on its content
  • Example: A classifier trained to identify "Hot Dog" vs. "Not Hot Dog" will label the entire image based on whether it contains a hot dog

Object Detectors

  • Choose when: You need to locate specific objects within images
  • What it does: Finds objects within images, draws bounding boxes around them, and applies appropriate labels
  • Example: An object detector trained on hot dogs will draw a bounding box around any hot dog in an image and label it specifically, rather than labeling the entire image

Next Steps

After identifying your use case and selecting the appropriate architecture: