Explore

Team Activity AI Architecture and Model Engineering Simulation

Here's a team activity designed for groups of 4 students to reinforce the concepts of AI architecture and AI model build engineering, emphasizing the connections between them.

---

# Team Activity: AI Architecture and Model Engineering Simulation

## Objective: Simulate the process of designing and building an AI language model, emphasizing architectural decisions and their impact on the engineering process.

## Materials: - Large sheets of paper or whiteboard - Colored markers - Sticky notes - Dice (one per team) - Timer

## Team Roles: 1. Architect 2. Data Engineer 3. Model Engineer 4. Project Manager

## Activity Structure: The activity is divided into 4 rounds, each representing a phase in the AI model development process.

### Round 1: Architecture Design (15 minutes)

1. The Architect leads this round. 2. The team must design a high-level architecture for a language model on the large paper/whiteboard. 3. They must include and connect these components: - Input Processing - Embedding Layer - Encoder (or Transformer) Blocks - Decoder (if applicable) - Output Layer

4. The Architect rolls the dice to determine a special constraint: - 1-2: Low computational resources - 3-4: Requirement for multilingual support - 5-6: Need for real-time inference

5. The team must adapt their architecture to accommodate this constraint.

### Round 2: Data Pipeline Design (10 minutes)

1. The Data Engineer leads this round. 2. Using sticky notes, the team must design a data pipeline that includes: - Data Collection - Data Cleaning - Tokenization - Data Augmentation - Batching

3. The Data Engineer rolls the dice for a data challenge: - 1-2: Limited dataset - 3-4: Noisy data - 5-6: Imbalanced classes

4. The team must modify their pipeline to address this challenge.

### Round 3: Model Implementation Planning (15 minutes)

1. The Model Engineer leads this round. 2. The team must create a pseudo-code outline for implementing their model, including: - Model class structure - Forward pass logic - Loss function - Training loop

3. The Model Engineer rolls the dice for an implementation challenge: - 1-2: Memory constraints - 3-4: Need for model interpretability - 5-6: Requirement for transfer learning

4. The team must adjust their implementation plan to meet this challenge.

### Round 4: Project Management and Integration (10 minutes)

1. The Project Manager leads this round. 2. The team must create a project timeline that integrates all previous elements: - Architecture design - Data pipeline implementation - Model development - Training and evaluation - Deployment

3. The Project Manager rolls the dice for a project constraint: - 1-2: Tight deadline - 3-4: Limited team resources - 5-6: Requirement for extensive documentation

4. The team must adapt their project plan to this constraint.

## Conclusion and Presentation (10 minutes)

Each team presents their final integrated plan to the class, explaining: 1. Their initial architecture and how it evolved with constraints 2. Key decisions in their data pipeline 3. Highlights of their model implementation strategy 4. How they integrated everything into a coherent project plan

## Debriefing:

After all presentations, discuss as a class: 1. How did architectural decisions impact data and model engineering choices? 2. What were the biggest challenges in integrating different components? 3. How did project management constraints affect technical decisions? 4. What strategies were most effective in adapting to unexpected challenges?

---

This activity reinforces several key concepts:

1. The interconnectedness of architecture, data engineering, and model implementation 2. The impact of constraints on design and engineering decisions 3. The importance of adaptability in AI project development 4. The role of project management in integrating technical components

By simulating the entire process from architecture to project planning, students gain a holistic understanding of AI model development and the crucial connections between different aspects of the process.

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.