04-652   Artificial Intelligence System Design

Course description

In a world where classical approaches increasingly fall short of the demands placed on modern intelligent systems, responsible AI-driven solutions have become essential. Imagine taking a real-world problem and turning it into a working AI system that is not only effective but reliable, scalable, and responsible. This course takes students on that journey, starting with the theory behind machine learning, deep learning, and data workflows, and guiding them through the principles of model development, evaluation, and deployment. Along the way, students learn how to design data pipelines, select appropriate models, and anticipate system behavior in practice. By the end, they are equipped to create AI solutions that are technically sound, ethically responsible, and ready to make an impact.

Students will also develop the ability to make informed design choices, understanding when to deploy lightweight models for efficiency and speed versus complex, heavy models for higher accuracy or richer capabilities. They will learn to weigh trade-offs in computation, latency, scalability, and resource constraints, ensuring that every AI system they design is not only effective but also practical and optimized for its intended context.

Learning objectives

By the end of the course, students will be able to understand the foundational theories and principles of machine learning, deep learning, and AI workflows, analyze and define business and technical problems to determine appropriate AI solutions, design and implement data pipelines and model development processes for end-to-end AI systems, and select and apply suitable machine learning and neural network models while balancing performance and efficiency. Furthermore, they will be able to evaluate, deploy, and maintain AI systems responsibly, ensuring reliability, ethical integrity, and scalability.

Outcomes

Upon completing this course, students will be able to:

• Explain the end-to-end AI system workflow, from data collection to deployment and monitoring.
• Build and evaluate both classical ML and deep learning models for real-world tasks.
• Design AI systems that are optimized for performance, efficiency, and resource constraints.
• Apply responsible AI principles to ensure fairness, transparency, and ethical deployment.
• Present a complete AI system design, justifying choices of models, architecture, and workflows.

Course content overview

This course is organized into seven (7) modules, progressing from foundational AI system concepts to deployment and responsible AI practices.

Module 1: Foundations of AI System Design

• Overview of AI systems and their components
• AI problem framing and system requirements
• Principles of reliability, scalability, and maintainability

Module 2: Data Pipelines & Feature Engineering

• Data collection, cleaning, and preprocessing
• Feature engineering and transformation
• Ensuring data quality, consistency, and representativeness

Module 3: Machine Learning Principles

• Supervised and unsupervised learning
• Classical machine learning algorithms and evaluation metrics
• Bias–variance trade-offs and model selection

Module 4: Deep Learning & Neural Networks

• Neural network fundamentals and training principles
• Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and embedding-based models
• Optimization, regularization, and generalization

Module 5: Recommender Systems & Applied Case Studies

• Collaborative filtering, content-based, and hybrid approaches
• Designing AI pipelines for recommendation tasks
• Evaluating system performance and user impact

Module 6: Deployment, Serving, and System Architecture

• Model serving patterns: batch, real-time, and streaming
• APIs, microservices, and lightweight containerization with Docker
• Monitoring, drift detection, and continuous system improvement

Module 7: Responsible and Reliable AI

• Fairness, transparency, and explainability
• Robustness, risk mitigation, and ethical considerations
• Balancing lightweight vs heavy models and resource trade-offs

Prerequisites

• Basic Python programming (including NumPy and Pandas)
• Comfort working with data
• Foundational mathematics:
• Linear algebra
• Calculus fundamentals
• Basic probability and statistics