Exercise: Module 1, Lesson 2 - Deconstructing the AI Stack

Objective: To apply the "AI Stack" model to a real-world AI-powered product or service, deconstructing it to identify the different layers and understand how they work together to create value.

Your Task

1. Select an AI-Powered Product: Choose a popular, consumer-facing application that clearly uses AI as a core part of its functionality. Examples include:

   • Spotify (for its recommendation engine)
   • Netflix (for its content recommendations and personalized artwork)
   • Grammarly (for its writing assistance)
   • TikTok (for its "For You" page algorithm)
   • A new AI-native application you are familiar with.

2. Deconstruct the Stack: For your chosen product, analyze and describe the three layers of its AI Stack. You will need to make some educated assumptions, as the exact internal details are often proprietary.

   • Hardware Layer: Who likely provides the massive computing power needed to train and run their AI models? (Hint: Think about major cloud providers and their relationships with chip makers).
   • Software Layer: What kind of foundation models might they be using? Are they likely using a model from a major provider (like OpenAI, Google), or have they built their own custom models, or a combination?
   • Application & Data Layer: This is the most important part.
     • What is the specific application of AI that the user experiences? (e.g., "Personalized song recommendations").
     • What is the crucial proprietary data that makes their AI effective and gives them a competitive advantage? Be specific. What data do they have that their competitors don't?

Deliverable

Write a short (300-500 word) analysis in a Markdown file. Structure your analysis with the following headings:

• Product Analyzed: [Your Chosen Product]
• Hardware Layer Analysis: [Your description]
• Software Layer Analysis: [Your description]
• Application & Data Layer Analysis: [Your description, with a focus on the proprietary data moat]

Example Submission Snippet:

Product Analyzed: Netflix

Hardware Layer Analysis:

Netflix runs its operations on Amazon Web Services (AWS). Therefore, the hardware layer consists of AWS data centers, which in turn use GPUs likely sourced from NVIDIA to handle the massive computational load of training and running Netflix's recommendation models for hundreds of millions of users.

Software Layer Analysis:

Netflix has a long history of building its own custom machine learning models. While they may use some off-the-shelf libraries, their core recommendation engine is a highly specialized, proprietary software asset. They are not simply calling a generic OpenAI API; they have engineered a custom solution tailored to media recommendation.

Application & Data Layer Analysis:

The application is the personalized homepage, content rows ("Because you watched..."), and even the custom artwork shown for each title. The proprietary data moat is Netflix's massive, historical dataset of user interactions: every show watched, every search query, every pause and rewind, ratings given, and time spent watching. This unique dataset, collected over more than a decade from a global user base, is an asset no competitor can replicate. It is the fuel that allows their custom models to generate highly effective, personalized recommendations.