Course Catalog

AGT-101: Foundations of AI Agents: Tools, Memory, Planning

Course Description

An introduction to autonomous AI agents. Covers the conceptual architecture: what makes a system an "agent" rather than a simple LLM call, how tools extend what agents can do, how memory allows agents to operate beyond their context window, and how planning enables multi-step tasks. Students build their first agent applications using the OpenAI and Anthropic APIs.

AGT-201: Building with the Claude Agent SDK

Course Description

A deep-dive into Anthropic's Claude Agent SDK. Covers tool definition, context management, streaming, structured output, and production patterns for building Claude-powered agents. Students build progressively more complex agents: a document Q&A agent, a web research agent, and a multi-step task automation agent.

AGT-210: Model Context Protocol: Building and Consuming MCP Servers

Course Description

The Model Context Protocol (MCP) is an emerging open standard for connecting AI agents to tools and data sources. This course covers MCP architecture: servers, transports (stdio and SSE), resource and tool definitions, authentication, and the Claude Desktop integration. Students build MCP servers for databases, APIs, and file systems, and consume MCP in agent applications.

CV-401: Image Formation and Classical Computer Vision

Course Description

Foundations of image formation: camera models, perspective projection, lens distortion. Classical image processing: filtering, edge detection (Sobel, Canny), corner detection (Harris, FAST). Feature descriptors: SIFT, SURF, ORB. Optical flow: Lucas-Kanade and Farneback methods. Stereo vision and disparity. The course provides essential background for understanding why deep learning methods work the way they do.

CV-410: Convolutional Neural Networks for Vision

Course Description

Deep learning for image understanding. CNN architectures: AlexNet through EfficientNet and ConvNeXt. Object detection: YOLO family, Faster R-CNN, DETR. Image segmentation: FCN, U-Net, Mask R-CNN. Transfer learning and domain adaptation. Data augmentation strategies. The shift from CNNs to ViTs and hybrid architectures. Students implement and train detection models on custom datasets.

CV-420: Vision Transformers and Foundation Models

Course Description

The vision transformer revolution. ViT, DeiT, Swin Transformer: patch embedding, positional encoding for 2D, hierarchical designs. CLIP: contrastive pretraining, zero-shot classification, image-text alignment. DINO and DINOv2: self-supervised learning with distillation. Segment Anything Model (SAM): promptable segmentation, architecture, zero-shot transfer. Florence, InternViT, and other vision foundation models.

CV-460: Vision-Language Models and Multimodal AI

Course Description

Multimodal systems that process both images and text. CLIP, ALIGN, and contrastive VLMs. LLaVA, InstructBLIP, and instruction-following VLMs. GPT-4V, Claude 3, and commercial multimodal APIs. Architecture patterns: early vs. late fusion, cross-attention, projection layers. Applications: visual QA, document understanding, image captioning, chart analysis. Evaluating multimodal models: benchmarks and failure modes.

ETH-401: Foundations of AI Ethics

Course Description

Philosophical and practical foundations of AI ethics. Value alignment as a technical and social problem. Moral frameworks: utilitarian, deontological, virtue ethics applied to AI systems. AI rights, moral status, and personhood debates. Historical parallels: prior technology transitions and their social consequences. Stakeholder analysis in AI systems. The difference between AI safety (alignment) and AI ethics (societal impact). Case studies: facial recognition, predictive policing, content moderation, hiring algorithms.

ETH-410: Algorithmic Fairness: Metrics and Mitigation

Course Description

Technical approaches to fairness in machine learning. Protected attributes and anti-classification. Statistical fairness definitions: demographic parity, equalized odds, equal opportunity, calibration. The impossibility theorems: why you can't satisfy all definitions simultaneously. Group vs. individual fairness. Fairness through unawareness and why it fails. Pre-processing, in-processing, and post-processing mitigation techniques. Fairness in foundation models and LLMs. Lab: audit a real credit scoring model using fairness metrics.

ETH-420: AI Regulation: EU AI Act and Global Frameworks

Course Description

The global landscape of AI regulation. EU AI Act: risk-based classification, prohibited AI practices, high-risk requirements, GPAI model obligations, enforcement timelines. US Executive Orders and NIST AI RMF. UK pro-innovation approach and AI Safety Institute. China's generative AI regulations and algorithmic recommendation rules. Sector-specific regulation: FDA guidance on AI medical devices, CFPB on credit algorithms, EEOC on hiring AI. Students analyze compliance requirements for a case-study AI product across jurisdictions.

ETH-430: Interpretability and Algorithmic Auditing

Course Description

Technical methods for understanding AI system behavior. Feature importance: SHAP, LIME, gradient-based attribution. Mechanistic interpretability: circuits, attention head analysis, activation patching. Probing classifiers. Concept activation vectors (TCAV). Auditing methodologies: internal audits, third-party audits, regulatory audits. How to audit an LLM: behavioral testing, red-teaming, capability elicitation. Documentation standards: model cards, datasheets for datasets, transparency reports.

GEN-101: Diffusion Models: Theory and Practice

Course Description

From first principles to working image generation. The forward diffusion process (adding noise), reverse denoising, DDPM, DDIM sampling. The score function and score matching. Latent diffusion: encoding to latent space, UNet denoising, decoding. CLIP text conditioning. Classifier-free guidance: intuition and implementation. FLUX architecture: flow matching and improved efficiency. Students generate their first images and analyze quality metrics.