Course Syllabi

Track Overview

Track 1: Undergraduate Engineering (one semester)

Audience. Third-year computer science or software engineering majors who can write Python comfortably, have seen basic linear algebra, and have completed at least one prior systems or web course. No prior machine learning or NLP background is assumed.

Goals. By the end of the semester, students can call frontier-model APIs, design effective prompts, build a working retrieval-augmented generation (RAG) pipeline over a corpus they choose, and deploy it behind a streaming endpoint with basic evaluation in place.

Prerequisites. Python 3.10+, Git, basic command line, basic linear algebra (matrix multiplication, dot products). The book's Section 5.1 (Platforms) and Appendix A cover gaps.

Capstone. Build a RAG application over a domain corpus the team chooses (lecture notes, a textbook, an API documentation set, a regulatory dataset). Required components: chunking strategy, embedding pipeline, retrieval with re-ranking, grounded generation, basic evaluation harness with at least 30 held-out questions, and a deployed streaming endpoint. Graded on correctness, faithfulness to the corpus, and engineering quality.

Track 2: Undergraduate Research (one semester)

Audience. Honors or research-track undergraduates who plan to pursue graduate study or research internships. Comfortable with mathematical notation and willing to read papers.

Goals. By the end of the semester, students can read and critique a 2024 to 2026 LLM paper, replicate a published result on accessible hardware, and explain the mechanisms behind their reproduction at the level of attention patterns or training dynamics.

Prerequisites. Same as Track 1 plus a prior course in machine learning, statistics, or signals.

Capstone. Replicate a published 2024 to 2026 LLM paper (with instructor approval) at a scale that fits the available compute budget. Deliverables: code, a 5-page write-up with figures matching the original where applicable, and a list of any deviations from the paper. Graded on faithfulness to the original methodology and on the depth of analysis of any discrepancies.

Track 3: Graduate Engineering (two semesters)

Audience. First-year M.Sc. students or industry-track Ph.D. students who already have some ML background and want to ship production LLM systems.

Goals. By the end of the year, students can train and adapt models, build agent systems, design evaluation harnesses, and operate production LLM services with monitoring, cost controls, and safety guardrails.

Prerequisites. Track 1 content (or equivalent: prior ML course plus production engineering experience).

Capstone. Multi-agent system with retrieval-augmented generation, an evaluation harness running in CI, deployment with monitoring, cost caps, and a rollback path. Graded on the production-readiness of the entire pipeline.

Track 4: Graduate Research (two semesters)

Audience. Research-track M.Sc. students or first-year Ph.D. students aiming for top-tier conference publications.

Goals. By the end of the year, students have a frontier-topic research project under way, with a written proposal, an experimental plan, baseline results, and the start of a workshop-paper draft.

Prerequisites. Track 2 content (or equivalent: comfortable with reading and writing ML papers).

Capstone. Original research project on a frontier topic. Required deliverables: a workshop-paper-quality draft (8 pages), reproducible code, and a 20-minute talk.

Track 5: Professional Bootcamp (10 weeks intensive)

Audience. Practicing software engineers and ML engineers transitioning into LLM work. Expect 25 to 30 hours per week of part-time engagement, structured around a working capstone.

Goals. By the end of 10 weeks, participants ship a production-ready agentic application with an evaluation harness, monitoring, cost controls, and a written architecture document explaining the model, infrastructure, and risk choices.

Prerequisites. Three or more years of professional software engineering. No ML background required.

Capstone. Production-ready agentic application in a domain of the participant's choice. Required: agent with tools, retrieval over a real corpus, evaluation harness, monitoring, cost controls, and a written architecture document covering model selection, infrastructure, and risk decisions. Graded on whether the system would survive a real production launch.

Reference appendices used across tracks

For all five tracks, the following appendices serve as on-demand references rather than required reading:

• Appendix A (Mathematical Foundations) and Chapter 0 (ML & PyTorch Foundations) for foundational gaps.
• Section 5.1 (Platforms) (Python for LLMs), Section 5.2 (Library Catalog), and Section 5.3 (Scripting Patterns & Environment Setup) for the toolchain.
• Hugging Face Transformers Deep Dive and LangChain Deep Dive for ecosystem familiarity.
• Reading Pathways for goal-based routes when a student wants to explore beyond the syllabus.