CS357: Foundations of Artificial Intelligence

[ Course Info | Course Objectives and Goals | Resources | Instructors | Textbooks | Schedule | Grading | Course Policies | Course Details ]

Welcome to CS357!

Course Info

• Course Number and Title: CS357 - Foundations of Artificial Intelligence
  • Section A
• Credit Hours: 4 Semester Hours
• Credit Hours include "contact time" in the classroom and outside course work. It is expected that the sum of classroom time and outside course work time should add up to three times the listed credit hours per week.
• Course Webpage: https://BillJr99.github.io/Ursinus-CS357-Fall2026


• LMS (Canvas): Canvas


• Course Calendar: Import the course calendar into your favorite calendar app with this link!


• Academic Term: Fall 2026-27


• Term Start and End: through


• Course Prerequisites: Prerequisites: CS-170 or CS-173 or DATA-201, or permission from the instructor.


• Class Meeting Locations and Times:
• Section A:
• s from 12:00 PM to 1:15 PM in Pfahler 107
• s from 12:00 PM to 1:15 PM in Pfahler 107
• Course Description: A technical introduction to the tools and practices that have evolved from artificial intelligence (AI) and large language models that considers the role of technology through the lens of the Ursinus Questions. This course prepares students to create and interact with conversational virtual assistants and multi-agent systems that achieve goal-oriented outcomes. Students will explore the foundations and history of artificial intelligence that enabled the technical underpinnings of generative AI, with a particular emphasis on the ethical, social, and intellectual property implications of the nature of the training data sets used to form large language models. The course includes a practical discussion of responsible AI from the perspective of creators and of consumers and stakeholders. Students will apply current design patterns in AI with respect to large language models to create and present a technical project using generative AI, and will lead a discussion focused on practices for responsible AI within the context of their chosen project. Prerequisites: CS-170 or CS-173 or DATA-201, or permission from the instructor. Four hours per week. Four semester hours.

Course Learning Objectives and Learning Goals

Learning Objectives

1. Analyze the ethical implications of AI agents acting on behalf of people, using case studies, governance frameworks, and design justice principles.
2. Design agentic AI systems, including prompts, retrieval pipelines, tools, and multi-agent orchestrations, that augment human capabilities while adhering to ethical guidelines.
3. Evaluate the reliability of AI agents across tasks and disciplines, identifying best practices, failure modes, and risks associated with their use.
4. Create documentation, evaluations, and governance plans for the responsible deployment of agentic AI technologies.
5. Evaluate the environmental impact of AI deployments and propose design choices that reduce energy and carbon costs without sacrificing core capability.
6. Engage with the philosophical and psychological dimensions of artificial intelligence, analyzing questions of understanding, responsibility, anthropomorphism, and the ethics of machine agency.

Learning Goals

1. Implement the perceive, plan, act loop of an agent using a locally hosted language model.
2. Explain how tokens, embeddings, attention, and sampling parameters shape the behavior of generative models, and verify these explanations with by-hand calculations.
3. Construct a retrieval-augmented generation pipeline over a personal knowledge base, and evaluate its retrieval and grounding quality.
4. Compose multi-agent systems using orchestration patterns including pipelines, critique and refine, debate, and stochastic consensus, while keeping each agent's context window small and focused.
5. Evaluate agent outputs using human rubrics and LLM-as-judge pipelines, and articulate the limits of each.
6. Author a governance and responsible-use analysis for an agentic system of your own design, and present it to a public audience.
7. Apply containerization and filesystem isolation principles to deploy AI agents with defined trust boundaries, non-root execution, read-only mounts, and minimal blast radius.
8. Design, implement, and secure a working MCP server with OAuth 2.0 client credentials flow, and connect it to a local agent to demonstrate tool discovery, invocation, and token lifecycle management.
9. Analyze the carbon and water footprint of an AI deployment and propose concrete efficiency improvements grounded in right-sizing, caching, and local-first inference choices.
10. Use a coding agent (OpenCode, Plandex, or Claude Code) to implement a feature from a written spec, then critique the generated diff for correctness, security, and test coverage.
11. Identify and research an issue, question, or practical problem
12. Develop a multi-disciplinary understanding of the problem to explore how it could be addressed
13. Collaborate to develop a strategic intervention that constructively addresses the issue
14. Communicate effectively with a variety of audiences through multiple modalities
15. Use the Open Questions to assess the learning process throughout, describing new understandings and specific areas of growth and skill development

The Questions

Throughout the course, we will be thematically guided by the Ursinus Questions:

• What should matter to me?
• How should we live together?
• How can we understand the world?
• What will I do?

This semester, we organize the course around a single guiding idea: an agent is a system that perceives, plans, and acts toward a goal. We will collectively consider questions like:

• What does it mean for a machine to act on our behalf, and how does delegation to an agent differ from using a tool?
• How do the foundations of generative AI (embeddings, attention, retrieval) enable, and limit, what agents can reliably do?
• How do the assumptions embedded in training datasets influence the behavior of agents, and what responsibilities do creators and users have in mitigating these effects?
• When should an agent run locally and privately, and when is it appropriate to rely on hosted services?
• How can multi-agent designs (debate, critique and refine, consensus) improve reliability, and what new failure modes do they introduce?
• How can system design align agent outcomes with ethical principles, and who is accountable when an autonomous system errs?
• How do coding agents like OpenCode, Plandex, and Hermes reshape software development, and what review discipline do their generated diffs demand?
• What is the carbon and water cost of AI computation, and how should environmental impact influence our design choices for agents?
• Does it matter whether an AI system genuinely understands, or only whether it behaves as if it does — and what are the ethical implications of our answer?
• How do containers, filesystems, network policies, and OAuth scopes create the safety boundaries that allow us to deploy autonomous agents responsibly?

Resources

Accommodations

Ursinus College and your instructor are committed to ensuring equal access and providing reasonable accommodations for all students. If you have, or think you have, a disability in any area such as, mental health, attention, learning, chronic health, sensory, or physical, please contact the Director of Disability Services.

As the instructor of this course, I strive to provide an inclusive learning environment. If you experience barriers to learning in this course, do not hesitate to discuss them with me. The Office of Disability and Access works with students who have any kind of disability, whether apparent or non-apparent, learning, emotional, physical, or cognitive, and need accommodations to increase their access to this learning environment. Students are encouraged to reach out to the disability and access team to discuss supports and accommodations they may need by scheduling a meeting using their scheduling link: https://kderstine.youcanbook.me/, or by emailing them at disabilityandaccess@ursinus.edu. Students can also review accessibility and disabilities services online at the Disability and Access at Ursinus Webpage. Their office is located in Lower Wismer, with the Institute for Student Success (ISS) office.

Peer Coaching

The Institute for Student Success offers Peer Coaching that you can sign up for anytime. The Institute for Student Success (ISS) is located in Lower Wismer, and connects students to the resources, activities, services, and programs that can help students be successful, thrive, and persist to graduation. They offer academic skills workshops, one-on-one coaching, tutoring, and more. Specifically, they offer course level tutoring as well as peer academic coaching (for help with time management, SMART goal setting, breaking down large assignments, and more). Contact them at instituteforstudentsuccess@ursinus.edu or 610-409-3400.

Wellness Center

If you are struggling with mental, physical, or substance use concerns that are negatively impacting your life, relationships, or academics, please reach out for support. The Wellness Center's Counseling staff and the Health Promotion staff collaborate to support students and they are all co-located in The Hive. The Wellness Center offers counseling and medical services at no cost to students. All services are confidential. For urgent mental health issues, a walk-in crisis hour is available at 2-3 pm each weekday, where students in crisis can be seen immediately by a clinician. A 24/7 on-call clinician is also a part of the campus Crisis Response Team. Health Promotion offers support services for students in recovery and Allies of Recovery training for friends/loved ones, and hosts many events and programs to help students build a healthy lifestyle and discover coping strategies. Visit ursinus.edu/wellbeing for wellness services campus wide and visit the office's website (https://www.ursinus.edu/offices/wellness-center/) for a full description of all services.

Center for Writing and Speaking

The Center for Writing and Speaking is available for one-on-one and group appointments to advise you as you revise your writing projects and presentations. They will even support your personal projects and extracurricular activities! Please feel free and encouraged to review any and all writing and speaking work from this class with them. Make an appointment at https://ursinus.mywconline.net/.

Bear2Bear and Basic Needs/Wellbeing

The college recognizes that temporary financial hardships can impact students' access to course materials, as well as their access to opportunities on campus. Please be aware of the Bear2Bear fund, which has been established by donors to the college and provides special grants for students who have exhausted other sources of funding. For information on assistance in getting other basic needs met, please visit https://www.ursinus.edu/wellbeing.

Help Room

The Math Help Room (Pfahler 102) is a great place to go if you are struggling and is managed by the Institute for Student Success. Students who have previously taken the course will be there to help you with the assignments.

Course Management Systems: Canvas, Microsoft OneNote, and Microsoft Teams

We will be using Canvas to post all of the grades. For the most part, we will submit work using Canvas as well. For class activities and notes, we will be using OneNote, and for other discussions and announcements for the course, including messaging me directly with questions, we will use Microsoft Teams. OneNote and Teams are linked to your Office suite through Ursinus, so you are automatically enrolled. There you can ask and answer questions about the lecture content and assignments.

Since it is likely that students will have similar questions, it is much more efficient for me to answer them there so the whole class can see the answer, so it is possible that I will ask you to re-send a question publicly that I get in an e-mail. If you'd prefer, I could anonymize the question as well, but I'd like you to have the opportunity to post it so that you are credited with having such a good question!

Course Instructors and Student / Office Hours

Role	Name and Contact Information	Student Hours / Office Hours
Professor	William Mongan Phone: 610-409-3268 E-Mail: wmongan@ursinus.edu Office: Pfahler Hall 101L	s from 11:20 AM to 11:50 AM in Pfahler Hall 101L s from 11:20 AM to 11:50 AM in Pfahler Hall 101L s from 3:00 PM to 5:00 PM in Pfahler Hall 101L s from 3:00 PM to 5:00 PM in Pfahler Hall 101L s from 3:00 PM to 5:00 PM in Pfahler Hall 101L Sign up for a one-on-one appointment during my office hours!

Student hours (also known as "Office Hours") are certainly available for asking questions about the course, about your assignments, and other academic questions you may have. You do not need an appointment to come to student hours, and you do not even need to have an agenda or set questions! You can come and just have a general chat about things with us. If you cannot make it to student hours, you can contact us for an appointment as well. Student hours are also for non-instructional topics of conversation: you can talk with me about your adjustments to college life, your long-term goals, advice about your academic journey, and most other things. If I don't know the answer to something or if I don't feel I am the best person to offer you advice about it, I will be happy to help connect you with the right people. In other words, don't be afraid to ask me things that you think are "off topic" - I love teaching because I love to be a resource for you on your journey. I'll be happy to see you there.

Textbooks

Required?	Title	Author	Edition	ISBN	Freely Available?
Required	Artificial Intelligence: A Guide for Thinking Humans	Melanie Mitchell	Revised Edition	978-1-250-40485-5
Required	AI by Hand	Tom Yeh			Online Version
Recommended / Supplemental	The Philosophy of Artificial Intelligence	Margaret A. Boden, editor		978-0198248545

Please Note: The cost of the book may be prohibitive for some students, so please note that renting the book is much cheaper. Please communicate as early as possible if you are having trouble obtaining the book, rather than keeping this to yourself, so that we can work on a solution together. If you are experiencing a financial hardship, please consider the Bear2Bear Emergency Fund for temporary relief applications.

Course Schedule

Week	Date	Title	Readings	Deliverables Handed Out	Deliverables Due
Week 1		Welcome: What Is AI, and What Is an Agent?	Mitchell, Prologue and Chapter 1 Interactive (optional): The Token Prediction Playground — an unplugged, in-browser simulator showing what next-token prediction can and cannot do Optional: AI Capabilities and Limitations (Anthropic Skilljar course) — a short self-paced companion to today's Capabilities and Limitations Framework The Project Thread: how the semester-long team project unfolds (read by the end of week 1) Optional: AI for All — tools overview slide deck	Warmup Assignment Handed Out Reading Group Discussion Leader Handed Out (extra credit, sign up for a slot) Team Formation Survey Handed Out
Week 1		The Agent Loop: Perceive, Plan, Act	Mitchell, Chapter 2 Russell and Norvig, Artificial Intelligence: A Modern Approach (4th ed.), Chapter 2 — Intelligent Agents Supplemental Tutorial: Shell Fundamentals for Agent Supervision
Week 2		Prompt Engineering as Agent Design: Personas and System Prompts	Mitchell, Chapter 3 (on what models do and do not understand) Chalmers, David J. "What We Talk to When We Talk to Language Models" Activity: System Prompt Design Workshop (see Part 3 of Prompt Patterns Assignment) Supplemental Tutorial: Agentic CLI Tools (Claude Code, Codex, Gemini CLI, and Friends) Supplemental Tutorial: Coding Agents — OpenCode, Plandex, and Hermes Supplemental Tutorial: The Karpathy Coding Approach — Spec-First, TDD, and Supervised AI Development		Team Formation Survey Due
Week 2		Running Your Own AI: Ollama, OpenWebUI, and Private Local Models	Mitchell, Chapter 3 Reading Response / Discussion: post a short response before class — what changed once the model ran on your own machine? Project Thread: teams announced today; before next class, read the Team Charter and Norms activity Supplemental Tutorial: Docker from First Principles Supplemental Tutorial: Terminal and Filesystem Isolation for Agent Safety Supplemental Reference: The Local Model Landscape — Llama, Mistral, Phi, Gemma, and Friends Supplemental Tutorial: The Hardware Behind AI — GPUs, Quantization, and Running Models at the Edge Supplemental Tutorial: RESTful LLM Access — The /v1/chat/completions Paradigm, curl, and the OpenAI SDK	Lab 1: Your First Local Agent Handed Out	Warmup Assignment Due
Week 3		Why Different Answers Every Time? Sampling, Temperature, and Generation	Mitchell, Chapter 3 Jurafsky and Martin, Speech and Language Processing (3rd ed. draft), Chapter 3 — N-gram Language Models Activity: Temperature and Sampling Explorer Supplemental Activity: Shakespeare GPT — Character-Level Language Models from Scratch Supplemental Discussion Activity: Deterministic vs. Probabilistic Computing — Automation Bias and Why AI Outputs Are Not Ground Truth Supplemental Activity: Machine Learning Foundations — Supervised Learning, Bias-Variance, and Ensembles		Signed Team Charter Due (all members sign; assessed within Class Activities)
Week 3		Hallucinations and Evaluating Agent Outputs	Mitchell, Chapter 3	Stakeholder Brief Handed Out (Project Thread)
Week 4		Tokens and Embeddings: How Agents Represent Meaning	Mitchell, Chapter 2 Nielsen, Neural Networks and Deep Learning, Chapter 1 Supplemental Activity: LLM Pre-Training Internals — BPE Tokenization, Training Objectives, and Scaling Laws Supplemental Activity: From Text Generation to a Neural Network — Workflow, Forward Pass by Numbers, and Visualization Supplemental Synthesis: Anatomy of an LLM Request — One Prompt, End to End, by Hand (threads tokenize → embed → attention → FFN → logits → sample → one training step with a single running example)	Written Assignment: Prompt Patterns and AI by Hand Handed Out	Lab 1: Your First Local Agent Due
Week 4		Hands-On: The Local Agent Stack (Tiers, Ports, and Compose)	Optional Theory Deep Dive: Attention and Transformers, Conceptually and by Hand AI by Hand, Attention worksheets (optional) Worksheet: A Neural Network Forward Pass by Hand (printable) Supplemental Activity: From Text Generation to a Neural Network (companion to today's stack session) Supplemental Tutorial: Agentic OpenWebUI — Tools, Uploads, and Multi-Agent Workflows Supplemental Tutorial: Containerizing AI Systems — Safety, Isolation, and Trust Boundaries
Week 5		Retrieval-Augmented Generation with Chroma	Mitchell, Chapter 4 Reading Response / Discussion: post a short response before class connecting RAG to a corpus of your own Supplemental Tutorial: Vector Databases — How Agents Search for Meaning Supplemental Video: LangChain RAG from Scratch (YouTube) Supplemental Video: Advanced RAG Techniques (YouTube) Supplemental Video: RAG Pipeline Deep Dive (YouTube) Supplemental Tutorial: A Threaded Auto-Research Pipeline (parallel workers, local summarization, merged digest)
Week 5		RAG Quality: Chunking, Clustering, and Reranking	Mitchell, Chapter 4 Supplemental Tutorial: Fine-Tuning, RAG, and Prompting — Choosing the Right Approach Supplemental Activity: Synthetic Data — Using AI to Train AI Activity: RAG Chunking Strategy Comparison (see Hands-On section) Supplemental Tutorial: The LLM Wiki Pattern (Wikis versus RAG)	Lab 2: A RAG Knowledge Base of Your Own Handed Out	Written Assignment: Prompt Patterns and AI by Hand Due Stakeholder Brief Due (Project Thread)
Week 6		Memory and the Small Context Window Principle	Activity: Structured Peer Review — SQR Cards and Team Check-Ins (in class: Stakeholder Brief exchange) Supplemental Tutorial: An Obsidian Second Brain with Agent Sync Supplemental Tutorial: Syncing Obsidian to GitHub and Wiring AI Agents to Your Vault Supplemental Activity: Memory Types in Agents — Working, Episodic, Semantic, and Procedural	Literature Review Handed Out (Project Thread; individual annotations, then team synthesis)	Intra-Team Check-In 1 Due (structured form, private to instructor)
Week 6		Tool Use and Function Calling	Activity: Tool Calling and Function Schemas (see Hands-On section) Supplemental Tutorial: Structured Outputs — JSON Mode, Tool Schemas, and Output Validation Supplemental Tutorial: Building an AI Chess Coach — LLM API Calls in a Real Web App
Week 7		Connecting Agents to the World: MCP and APIs	Supplemental Tutorial: MCP Deep Dive — REST APIs, OAuth 2.0 Flows, and Secure Tool Invocation Supplemental Tutorial: GitHub Power-User Tools — gitingest, getmcp.io, deepwiki, gdagram, and github.dev		Lab 2: A RAG Knowledge Base of Your Own Due
Week 7		Design First: Plan Your Agent System Before You Build It	Supplemental Activity: Designing Agent Personas and System Prompts Supplemental Tutorial: Designing Your AI Development Environment — Context Models, AGENTS.md, and Superpowers Supplemental Tutorial: Agent Skills and Plugins — Writing, Configuring, and Publishing Skills for OpenCode and pi.ai	Written Assignment: Design Before You Build Handed Out (optional assignment)	Literature Review: Individual Annotated Bibliographies Due (Project Thread milestone)
Week 8		Advanced Agent Loops: Reflection, Recovery, and Control Flow	Russell and Norvig, Artificial Intelligence: A Modern Approach (4th ed.), Chapter 2 — agent architectures, from simple reflex to utility-based and learning agents
Week 8		Orchestration Patterns: Pipelines, Routers, and Planners	Supplemental Tutorial: Agent Frameworks — LangChain, CrewAI, AutoGen, and Agno Supplemental Tutorial: Cost Optimization — Token Budgets, Caching, and Model Routing Supplemental Activity: Debugging AI Agents — What Went Wrong and How to Find Out Supplemental Video: Agentic Workflows — a full free course on orchestration, multi-agent MCP, and agent chatrooms (Nick Saraev) Supplemental Case Study: Governing Coding Agents — Charters, Handoffs, and Durable Memory	Lab 3: Critique and Refine Handed Out	Literature Review: Team Synthesis Due (Project Thread)
Week 9		The Critique and Refine Pattern	Project Thread: midpoint Team Charter revisit (in class; bring your charter) Supplemental Tutorial: Human-in-the-Loop — Oversight, Escalation, and Appropriate Autonomy Supplemental Case Study: From Second Brain to Chief of Staff — A Personal Agent in Production	Written Assignment: Design Your Agent Operating System Handed Out (optional assignment)	Written Assignment: Design Before You Build Due (optional assignment)
Week 9		Multi-Agent Debate	Reading Response / Discussion: post a short response before class — when does a multi-agent design help, and when does it just cost tokens? Supplemental Activity: Multi-Agent Communication — Protocols, Shared State, and Coordination
Week 10		Stochastic Multi-Agent Consensus		Lab 4: Multi-Agent Debate and Consensus Handed Out	Lab 3: Critique and Refine Due
Week 10		Agent Teams: Specialists over Monoliths	Supplemental Tutorial: Publishing Artifacts (GHCR, Docker Hub, and npm) ShipIt Guide: Build, Test, CI, and Publish (required for Final Project) Supplemental Activity: Deploying Agents — From Laptop to Production Supplemental Tutorial: Scheduled and Autonomous Agent Runs — Cron, systemd, n8n, and GitHub Actions (clone → review → open a PR)	Final Project: Custom Agent Team Handed Out Final Project: Responsible AI Audit Handed Out (alternative track) Final Project: Build and Publish an Open-Source Agent Handed Out (alternative track)
Week 11		Visual Agent Building with Langflow	Activity: Building a Visual Agent with Langflow (see Part IV: Hands-On Langflow Build) Supplemental Tutorial: Cloudflare Workers and Pages with Wrangler Supplemental Tutorial: Multimodal Agents — Vision, Documents, and Code as First-Class Inputs Supplemental Video: n8n for Everyone — visual AI agents and workflows (Nick Saraev)		Intra-Team Check-In 2 Due (structured form, private to instructor) Written Assignment: Design Your Agent Operating System Due (optional assignment)
Week 11		Evaluating Agents: LLM-as-Judge and Rubric Pipelines	Supplemental Tutorial: Testing Agents — Evaluation, Regression, and the Non-Determinism Problem Supplemental Activity: Agent Observability and Distributed Tracing Supplemental Activity: Benchmark Design — How We Know If AI Systems Work Supplemental Activity: The Economics of AI Agents — Pricing, Incentives, and Market Dynamics Supplemental Activity: LLM Serving in Production — TTFT, KV Cache, Batching, and Cost Modeling Supplemental Activity: Integrating AI with Institutional Systems — LMS, Assessment Software, and Student Information Systems	Lab 5: An LLM Rubric-Grading Pipeline Handed Out	Lab 4: Multi-Agent Debate and Consensus Due
Week 12		Agentic Case Studies: Migration, Browsing, and Research Agents	Project Thread: cross-team Proposal peer review (in class; SQR protocol) Supplemental Tutorial: Agent Security — Threat Modeling and the OWASP LLM Top 10 Supplemental Tutorial: Prompt Injection — Attacks and Defenses Supplemental Activity: Adversarial Robustness and Red-Teaming LLMs Supplemental Activity: AI Coding Agent Security — Poisoned Repos, the Software Supply Chain, and State-of-the-Art Defenses (indirect injection via repo artifacts, slopsquatting/dependency confusion, spotlighting/CaMeL/instruction-hierarchy, and the lethal trifecta) Supplemental Activity: Red-Teaming and Adversarial Robustness — Prompt Injection, Jailbreaks, and PAIR Supplemental Activity: Using AI for Academic Research — Literature Search, Citation Checking, and Structured Workflows	Written Assignment: Does It Matter If Machines Understand? Handed Out (optional assignment) Written Assignment: Model Cards and Datasheets Handed Out (optional assignment)	Final Project Proposal Due
Week 12		Training Data and Bias	Reading Response / Discussion: post a short response before class — one design choice you would make differently in a system you deploy Coded Bias (film), watch before class O'Neil, Weapons of Math Destruction — the chapter on predictive models in education Benjamin, Race After Technology, Chapters 1–2 The Revenge of the Philosophy Majors (Wallace, The New York Times, 2026) Supplemental Case Study: Early-Alert and At-Risk Identification with Local AI (synthetic data) Supplemental Activity: The Philosophy and Psychology of Artificial Intelligence Supplemental Activity: Ethical Frameworks for Agentic AI Systems Supplemental Activity: Dataset Documentation — Model Cards and Datasheets Supplemental Activity: AI in Education — Opportunity, Disruption, and the Integrity Question Supplemental Activity: AI and the Future of Work — Automation, Displacement, and Adaptation Supplemental Activity: AI for Scientific Discovery — AlphaFold, Drug Discovery, and Reproducibility
Week 13		Intellectual Property, Privacy, and the Case for Local AI	Crawford, Atlas of AI, Chapter 6 — Affect / the politics of data Supplemental Activity: Privacy-Preserving AI — Federated Learning, Differential Privacy, and PII Scrubbing	Written Assignment: Governance and Policy Handed Out Lab 6: Responsible AI Handed Out	Lab 5: An LLM Rubric-Grading Pipeline Due
Week 13		Governance and Policy Writing	Christian, The Alignment Problem — any one chapter (e.g., the section on reward and reinforcement) Activity: AI Policy Writing Workshop (see Part 0 of Governance Assignment) Supplemental Activity: AI for Accessibility — Opportunity, Obligation, and Risk Supplemental Activity: Neuro-AI Ethics — Brain-Inspired AI, Cognitive Science, and the Study of Mind Supplemental Activity: AI Alignment and Safety — From RLHF to Constitutional AI Supplemental Activity: Reinforcement Learning and RLHF — How Human Feedback Shapes AI Behavior Supplemental Activity: What Kind of Model Is This? Base, Instruct, Reasoning, and Vision Models Supplemental Activity: AI Regulation — EU AI Act, NIST AI RMF, and Sector Rules	Written Assignment: Mapping a Real AI System to the Regulatory Landscape Handed Out (optional assignment)	Written Assignment: Model Cards and Datasheets Due (optional assignment)
Week 14		Explainability and Human-Centric Design	Carr, The Glass Cage, Chapters 2–3 — automation, deskilling, and automation bias Supplemental Tutorial: The AI Maker Module (Testing, CI, and Human-Centric Design) Supplemental Activity: Explainability in Depth — SHAP, LIME, Attention, and the Limits of Interpretation	Written Assignment: The Carbon Cost of Intelligence Handed Out (optional assignment)	Written Assignment: Does It Matter If Machines Understand? Due (optional assignment)
Week 14		Environmental Impact and the Carbon Cost of Intelligence	Crawford, Atlas of AI, Chapter 1 — Earth: the mineral and energy costs of computation Reading Response / Discussion: post a short response before class on the carbon and water cost of a deployment you would run Supplemental Activity: AI Creativity — Generative Models, Authorship, and the Nature of Originality		Written Assignment: Governance and Policy Due
Week 15		Project Studio and Gallery Walk	Project Thread: gallery-walk peer review (SQR protocol)		Written Assignment: Mapping a Real AI System to the Regulatory Landscape Due (optional assignment) Intra-Team Check-In 3 Due (structured form, private to instructor)
Week 15		Project Studio: Final Integration and Demo Rehearsal			Written Assignment: The Carbon Cost of Intelligence Due (optional assignment) Lab 6: Responsible AI Due
Week 16		Demo Day: Final Project Presentations (Class Switch Day: follows a Thursday schedule)			Final Project: Custom Agent Team Due Final Project: Responsible AI Audit Due (alternative track) Final Project: Build and Publish an Open-Source Agent Due (alternative track)

Please note the following holidays this term:

Please note the following key calendar dates:

• Add Deadline:
• Drop with a W Deadline:
• Reading Day:

Grade Breakdown

Letter grades will be assigned on the scale below at the end of the course. "Grade grubbing" is not conducive to professional practice; every assignment has or will have very precise expectations and point breakdowns, and I will evaluate submitted work carefully according to these standards. I will also return assignments in a timely manner, and the running weighted grades will be updated frequently. Therefore, I expect a commensurate level of respect from you. In sum, you should know where you stand at all times, there will be plenty of opportunities to improve your standing, and there should be no surprises at the end of the course.

Grading Table

Item	Weight
Labs (6 required, each with a direction you choose)	35%
Written Assignments (choose 3 from the 7 offered)	20%
Final Project and Project Thread milestones (choose one project: Custom Agent Team, Responsible AI Audit, or Open-Source Agent)	25%
Class Activities and Participation	10%
Reflection Notebook	10%

Letter Grades

Letter	Grade Range
A+	96.9-100
A	93-96.89
A-	89.5-92.99
B+	87-89.49
B	83-86.99
B-	79.5-82.99
C+	77-79.49
C	73-76.99
C-	69.5-72.99
D+	67-69.49
D	63-66.99
D-	59.5-62.99
F	0-59.49

Course Policies

Courtesy of the Online Education Blog of Touro College.

Classroom Environment and Inclusivity Standards

In this class we will work to promote an environment where everyone feels safe and welcome, even during uncomfortable conversations. As we explore these ideas, every voice in the room has something of value to contribute to group discussion. Every participant must show respect for all others. You are encouraged to not only take advantage of opportunities to express your own ideas, but also to learn from the information and ideas shared by other students. Participation is crucial to the success of this classroom experience. Your insights, questions and comments will be useful not only to yourself and to your a professor, but to your fellow students. My goal is to foster a environment in which students across all axes of diversity feel welcome and valued, both by me and by their peers. Axes of diversity include, but are not limited to, age, background, beliefs, race, ethnicity, gender/gender identity/gender expression (please feel free to tell me in person or over e-mail which pronouns I should use), national origin, religious affiliation, and sexual orientation. Discrimination of any form will not be tolerated. Please see me if you have any questions about our classroom environment. Discriminatory acts can be reported by following the process outlined on the college policy on discriminatory acts. Furthermore, I want all students to feel comfortable expressing their opinions or confusion at any point in the course, as long as they do so respectfully. As I will stress over and over, being confused is an important part of the process of learning computer science. Therefore, I will not tolerate any form of put-downs by one student towards another about their confusion or progress in the class. Learning computer science and struggling to grow is not always comfortable, but I want it to feel safe. Much of this material is probably new to everyone, and those with some prior experience likely recall a time when it was new to them, too. Remember that this is not a competition: helping others to grow is itself a richly rewarding professional development opportunity. In order to allow for equitable access to class for students who may be attending and participating remotely, I may record our class sessions. These recordings will only be available on our Canvas site. I will announce that we are recording in the beginning of any classes of this kind; out of respect and privacy for me and all class members, please do not download, copy, or redistribute class recordings.

Online Communication Policy

Since this is a class-wide communication, the following rules apply to message groups and electronic communications:

1. Students are expected to be respectful and mindful of the classroom environment and inclusivity standards.
2. They are equally applicable to a virtual environment as they are in class.
3. Students are not permitted to share direct answers or questions which might completely give away answers to any homework problems or labs publicly on Microsoft Teams. When in doubt, please send me a direct message there.
4. I will attempt to answer questions real time during student / office hours. Otherwise, I will make every attempt to respond within 24 hours. Of course, students can and should still respond to each other outside of these intervals, when appropriate!
5. Students may ask anonymous questions.

Early Alerts

From time to time, I may send academic early alerts through the college to you regarding your academic performance or engagement in the class. These alerts are intended to help you engage in a way that will improve your ability to be successful. Should I send you an alert, I expect that you will follow up with me within 5 days to discuss your engagement on campus or in our class.

Attendance Policy

Students may miss up to 4 classes without justification, although students are encouraged to communicate with me prior to missing class (or immediately after) so that we can discuss what was missed and how to catch up. Any student who misses more than 4 classes will receive a full letter grade reduction for each subsequent class missed from the final letter grade. A lateness to class shall count as one-half of an absence for purposes of this policy.

Religious and Spiritual Life Observance Policy

Per the Religious and Spiritual Life Observance Policy, students who expect to miss classes, examinations, or other assignments due to religious observance may notify me two weeks prior to the observance. I will be happy to discuss reasonable alternatives with you.

Collaboration Policy and Academic Integrity Policy

Communication between students is allowed (and encouraged!), but it is expected that every student's code or writeups will be completely distinct! Please do not copy code off of the Internet (repurposing code from the Internet will probably make it harder anyway because the assignments are so specialized). Please cite any sources in addition to materials linked from the course website that you used to help in crafting your code and completing the assignment.

See the Course Management page in the Faculty Handbook for an explanation of college policies on plagiarism and other academic honesty violations.

To encourage collaboration, students will be allowed to choose one "buddy" to work "near" during the assignment. Students are still expected to submit their own solutions, but they are allowed to provide substantial help to their designated buddy, and even to look at the buddy's code during the process. Students must indicate their buddies in the README upon assignment submission. Please let me know if you would like a buddy but are having trouble finding one.

Below is a table spelling out in more detail when and how you are allowed to share code with people (table style cribbed from Princeton CS 126).

Please Note: The terms "exposing" and "viewing" exclude sending or ingesting electronically, which would be considered copying. Exposing and viewing are normally done in the context of in-person working or in the help room. In addition, "Other People" includes internet sources!

	Your Buddy	Course Staff	Course Grads	Classmates	Others (Including Generative AI)
Discuss Concepts With	OK	OK	OK	OK	OK
Acknowledge Collaboration With	OK	OK	OK	OK	OK
Expose Your Code/Work/Solutions To	Labs Only	OK	OK	NO	NO
View the Code/Work/Solutions Of	Labs Only	OK	NO	NO	NO
Copy Code/Work/Solutions From	NO	NO	NO	NO	NO

If the work you submit appears to be copied from previous work or the collaboration policy has been violated in any way (including working with more collaborators or "buddies" than the course deliverable specifies) according to the College Academic Honesty policy, regardless of intent, then it may be an academic dishonesty case, and it will be referred to the Office of the Provost. I am required to make this report in every occurrence, so it is best to speak with me first if there are any questions about the policy or expectations. You should feel free to have these conversations with me anytime prior to making your submission without fear of penalty. Finally, except as specified in this collaboration policy, it is expected that your work is your original work. You must cite any collaborations or references that you use, including consultation with or work generated through the use of generative AI systems. You may have a friend or relative with computing experience, but they should not do your assignments, labs, etc., for you. Similarly, generative AI systems should not do your work for you. In general, consultation with AI systems should be treated as assistance from a classmate per the table above, and should be cited with both the prompt and the response. Should a question arise as to the level of support given by AI or an outside source that is fully cited by the student in the submission, the student will be invited to resubmit that work without such resources without constituting an academic integrity violation.

Flexible Submission Policy

In the absence of accommodations arranged in advance with the instructor or college, all assignments are due at 11:59 PM Eastern Time on the date(s) stated on the schedule. With prior permission and a reasonable first draft submission by the deliverable deadline, any student may request a three day extension on any deliverable, as often as needed. Assignments will be accepted without prior permission following the original deadline, or, if requested, following the three-day extension deadline, with a points deduction of 10% per day if submitted before 11:59 PM Eastern Time on the day submitted. If a student adds the course late, deliverables due prior to or on the day of that student's registration will be due twice the number of days following the first day of the semester that they registered (for example, a student who registers on the third day of the semester shall receive six days to submit assignments from the first three days, and then the remainder of this policy takes effect for those and for all other deliverables). Under no circumstances (including accommodations) can late work be accepted after the final class meeting, nor during final exams week, nor after the exam.

A Word About Submitting Work On-Time

Managing your time and pacing yourself consistently are crucial to your academic success. In professional practice – and in the spirit of the Ursinus Question “how should we live together?” – others will depend on you and will build upon the work you create. In the classroom, these interactions are modeled in the form of group projects and activities, and also in the form of cumulative course content that builds upon itself thematically throughout the semester. Research indicates that self-imposed or flexible deadlines does not lead to optimal scheduling [1], which, in turn, can lead to a compounding of overdue work across multiple classes. In addition, extensions to or prolonging of assignment deadlines has been shown to yield a detrimental appearance of complexity [2]: we tend to believe that assignments with longer durations are more difficult, and can find it more difficult to get started due to the anxiety that results. Your professor has established a schedule and procedure for completing and submitting classwork that complements the topics being covered during the semester. The specific details of that schedule and of those procedures may vary from instructor to instructor, depending upon the unique needs and instructional approach of the class. These details are specified in the course syllabus, and because those details have been designed thoughtfully and intentionally to best enable your consistent engagement with the class, the guidelines in that syllabus pertaining to engaging in the course, completing work, and posting grades (including a grade of incomplete) shall be considered effective policy for the course. Regardless of the implementation details from one course or from one instructor to another, these instructional designs are intended to enable you to engage with the course in a healthy and consistent manner, to manage your time effectively between your class, your other classes, and your extracurricular activities, and to better position you for success in class and beyond.

(References: [1] Ariely, Dan, and Klaus Wertenbroch. “Procrastination, Deadlines, and Performance: Self-Control by Precommitment.” Psychological Science, vol. 13, no. 3, May 2002, pp. 219–224, doi:10.1111/1467-9280.00441. [2] Meng Zhu, Rajesh Bagchi, Stefan J Hock, The Mere Deadline Effect: Why More Time Might Sabotage Goal Pursuit, Journal of Consumer Research, Volume 45, Issue 5, February 2019, Pages 1068–1084, https://doi.org/10.1093/jcr/ucy030.)

Grade Posting Policy

Feedback and grades will be provided frequently, generally within one week of the due date of any deliverable, and no more than two weeks following the due date. Inquiries seeking a change of grade must be made within 7 days of the posting of the grade in question, including the posting of a reduced grade for a missing submission. Final grades are due within 48 hours of the final exam (or last class scheduled meeting in a class with no final exam); grades are not subject to change (including a change from a posted grade to a grade of incomplete) once submitted to the college.

Incomplete Policy

A grade of I may only be granted by permission of the Office of the Provost. A petition to the Office of the Provost will only be made upon written request by the student, including all information requested by the Office of the Provost. Requests for a grade of I will only be made in situations where such a request is warranted. Specifically, the student's grade must be passing at the time the request is made. A petition for a grade of I will not be considered if an academic alert was submitted by the instructor prior to the course Drop with a W (withdraw) deadline.

Title IX

Title IX is a federal law, under which it is prohibited to discriminate, harass, or commit misconduct on the basis of gender or sex. The Title IX Coordinator is available to receive inquiries and to investigate allegations in this regard. As a professor, I am a mandatory reporter under Title IX, and am required to report disclosures made to me related to Title IX.

Inclement Weather and Class Cancellation Policy

In the event that the College closes due to inclement weather or other circumstances, our in-person class sessions, student / office hours, labs, or other meetings will not be held. I will contact you regarding our plan with regard to rescheduling the class or the material, any assignments that are outstanding, and how we can move forward with the material (for example, any readings or remote discussions that we can apply). If necessary, I may schedule online virtual sessions in lieu of class sessions, and will contact you with information about how to participate in those. I will communicate this plan to the department so that it can be posted on my office door if it is feasible to do so. This policy and procedure will also apply in the event that the College remains open but travel conditions are hazardous or not otherwise conducive to holding class as normal. Should another exigent circumstance arise (for example, illness), I will follow this policy and procedure as well.

Class Recording Policy

In order to allow for equitable access to class for students who may be unable to attend, I may record our class sessions. These recordings will only be available on our Canvas site. I will announce that we are recording in the beginning of any classes of this kind; out of respect and privacy for me and all class members, please do not download, copy, or redistribute class recordings. Please do not record classes without first discussing it with the instructor and, as appropriate, with appropriate accommodations to do so.

Student Perception of Teaching Questionnaire (SPTQ)

I will be soliciting student feedback through the SPTQ and possibly through other forms of commentary. This feedback greatly assists me and the department as we develop our courses and overall curriculum for this program. This course has benefitted from the feedback of those students who took the course before you, and your feedback will help maintain and improve the course for those to follow. I strongly encourage you to participate in this important and valuable process.

Syllabus Subject to Change

I will do my best to provide all relevant information about the course on this syllabus. Sometimes, exigent circumstances, the pace of the class, or other circumstances will warrant minor revisions to the syllabus. For example, inclement weather or other campus closure might affect the course schedule and assignment deadlines; in addition, I may find that the class benefits from spending more time on a particular topic, and adjust accordingly. Although I try to avoid rescheduling student / office hours, it may become necessary from time to time to accommodate other events in the College. Should any revisions be necessary, I commit to making any revisions in my estimation of the best interests of the class, and commit to communicating those changes to you as soon as I make them.

Course Details

This course is about building agents you understand and can run yourself. The sections below explain how the course gives you real choice over your path, how to read an assignment, how to prepare for each hands-on session, and how the day-to-day work of participating — including engaging with your classmates’ work — is valued and evaluated. Read them now, and return to the participation and preparation guides throughout the term.

How This Course Works: Choice and Universal Design

This course is designed as deliberate choice architecture, in the spirit of Universal Design for Learning: there are many routes through it, several ways to demonstrate what you have learned, and no path is the “remedial” one. You have real authorship over your semester.

• Choose your direction. Everyone completes the same 6 labs, but each lab offers multiple directions you can take the work — so you build the same core skill as your classmates, then extend it toward what interests you. You also choose 3 written assignments from the 7 offered and 1 final-project track from 3 (a Custom Agent Team, a Responsible AI Audit, or an Open-Source Agent). The hands-on and the analytical are equally valid ways to earn your grade; build the balance that fits how you learn.
• A shared spine. The semester-long Project Thread is the one path everyone walks together, so that individual choice never means working alone. It carries the team milestones — charter, stakeholder brief, literature review, proposal, and demo — and the peer review that ties the section together.
• Depth inside every lab. Each of the 6 labs opens with a shared core that everyone builds, then a menu of directions — local model internals, containerization, MCP and OAuth, coding agents, fine-tuning, observability, prompt-injection defense, privacy, explainability, and more. Pick the direction that pulls you; if the one you want is not on the menu, propose it. Supplemental tutorials and activities remain available throughout for going deeper where your interest leads.

If a path you want is not on the menu, propose it. The point of the choices is to let you leave this course able to stand up, operate, and reason about an AI system of your own.

How Assignments Are Structured: Purpose, Task, and Criteria

Every assignment is written to be transparent about three things, so you are never guessing about what is being asked or how it will be judged:

• Purpose — why the assignment exists and what capability it builds toward.
• Task — what you will actually do, in concrete steps.
• Criteria — how your work will be evaluated. Every graded assignment carries a rubric with four levels (pre-emerging, beginning, progressing, proficient), so you can see what proficient work looks like before you begin, and can hold your own draft against it.

Read the Purpose first: it tells you what the assignment is really for, which is the fastest way to make good decisions when the task gets open-ended — and much of this course is deliberately open-ended, because operating real systems is. Every assignment also asks you to reflect and to disclose your use of AI tools honestly; that reflection and that disclosure are part of the work.

Preparing for Each Class

Our meetings are hands-on POGIL sessions: you work in your team through activities that build the concepts and run the systems, not lectures you passively receive. Class works best when you arrive ready, and “ready” is a routine you can run rather than a matter of luck. The Preparing for Each Class guide lays it out: how to work through a technical activity in passes, how to attempt the reading responses beforehand, and how to arrive with a question or a stuck point — often something you tried to run on your own machine that did not behave. Bringing that is the accountability check that the preparation happened, and it is usually where the best discussion starts.

Class Activities and Participation (10%)

This is a course you do, not one you watch. This component values the daily work of showing up prepared, contributing to the shared build, and engaging seriously with your classmates’ work. It is assessed against the rubric on the Preparing for Each Class guide, across four dimensions: preparation, contribution, collaboration, and reflection. It takes several forms, by design:

• In-class activities. Your team rotates the POGIL roles — Manager, Recorder, Presenter, and Reflector — so that on different days you facilitate, capture the group’s thinking, report out, or synthesize. Posting your team’s answers to the class discussion board is participation the whole class learns from.
• Reading responses and discussion. From time to time the agenda sets aside time to discuss a reading or a result, prepared by a short reading response you write beforehand. These are marked on the schedule.
• The student-led Reading Group. When a classmate leads a Reading Group discussion, the audience has a job too: engaging with the presenter’s source and question is part of your participation grade, and the reading response guide explains the brief pre-read note or in-session question that earns it. (Leading a session remains separately available for extra credit.)
• Project-Thread peer review. The structured SQR peer reviews — one concrete Strength with evidence, one genuine Question, one Risk with a suggested mitigation — that you give other teams at the stakeholder-brief, proposal, and gallery-walk stages are participation of the most professional kind, and they count here.

Participation takes more than one form on purpose. If the spoken room is hard for you, the written channels — the discussion board, reading responses, and SQR cards — are real ways to earn this component. Being confused is part of learning this material; talk with me early and we will find the path that fits.

Reflection Notebook (10%)

This course asks not just how to build agents but whether and when we should, and your Reflection Notebook is where you think that through. Keyed to the four Ursinus Open Questions, it collects your responses to the reflection prompt that closes each activity, lab, and Project-Thread milestone. It is reviewed at midterm and at the end of the term; the guide explains what to keep and how it is evaluated.