Modern data systems are rapidly evolving to power the next generation of AI-driven and intelligent workloads. This seminar course covers recent developments from both industrial deployment and academic research. We will cover the latest trends in data processing on modern cloud architectures, recent work to expand the frontier of data processing to vectors and semantic operators, and experimental agentic search and deep research systems.

The course mixes two kinds of lectures: most meetings are topic lectures that discuss technical material from cloud-native data systems through the retrieval, semantic, and agentic substrate for AI workloads; there is also a mini guest-lecture series that runs through late October and early November featuring builders of emerging systems from both industry and academia.

Each topic lecture splits time between a short instructor lecture on the topic and a group discussion of the assigned readings. Come prepared to discuss each paper's story (its problem, key idea, and results) and offer critique of its weaknesses, unfair baselines, or missing experiments. Guest lectures have no assigned readings, but please come prepared to engage with the speaker.

Reading & Check-ins: All students are expected to read the assigned papers and are encouraged to write their own reviews or notes as they go. Rather than submitting reviews online, there will be six in-class check-ins on the dates marked in the schedule below: class opens with a 10-min closed-book quiz (no electronics) on a random paper from recent reading. Questions are intended to be general: if you read the paper and thought about it beforehand, you should find them straightforward. Your lowest check-in is dropped when computing the final score.

CS 564 or equivalent. If you have concerns about meeting the prerequisites, please contact the instructor. There is no formal textbook for this course.

The course uses Canvas for course management. If you are enrolled, you should already have access; if not, reach out to the instructor via email. Course materials, grades, and announcements will be posted on Canvas, so please check Canvas regularly or turn on notifications. Canvas is also where you will submit all project deliverables (team list, proposal, mid-term report, final report and code).

Here are some links to University policies and resources you may find helpful:

• Academic integrity
• Accommodations for students with disabilities
• Diversity and inclusion
• Religious observances
• Privacy of student records and lecture materials
• Students' rules, rights, and responsibilities
• Student health and well-being
• Teaching and learning data transparency
• Course evaluations

• Project proposal: 15%
• Mid-term report & presentation: 15%
• Final report & presentation: 40%
• In-class reading check-ins (pass/fail): 15%
• Class participation: 15%

Check-in & deadline policy: In-class check-ins are closed-book and held on the scheduled dates; there are no make-ups for an unexcused absence, but your lowest check-in is dropped when computing the final score. It's your responsibility to come to class on time so a to not miss a check-in. Please email the instructor ahead of time if you have an excused absence or otherwise cannot submit something on time due to extrenuous circumstances.

A big component of this course is the course project where you pick a topic in data management and explore it in depth. You are encouraged to select a project in the topics covered by this course, but projects outside of these topics are also welcome subject to instructor approval. The course project is a group project and you may have up to 3 students in a group (teams of 3 are preferred). Please start looking for project partners right away.

The project runs the full semester through a sequence of deliverables — team formation, a proposal, a mid-term written report, and a final presentation with final written report:

• Team formation (due Tue 9/15): submit one team list per group — member names and any preliminary thoughts on your topic. If you don't have a full team or a topic yet, say so and the instructor will help place you.
• Proposal (due Thu 10/8, with a 5 min talk at the proposal workshop): a short written document covering the project title and team; an abstract of the "big idea"; motivation (why it's interesting, how it compares to and advances existing work, and your hypothesis); evaluation and success metric; a rough plan detailing division of labor, project milestones, and timeline.
• Mid-term report (due Tue 11/10, with an 8 min in-class presentation on 11/5 or 11/10): a short write-up documenting your progress — completed vs. open tasks and deliverables, results so far, any problems and how you'll address them, an updated timeline with justification, and who did what. You can reuse parts of it in the final report.
• Final presentation (12/1–12/8): a 15 min in-class conference-style talk.
• Final report & code (due Wed 12/9 last day of classes): a conference-style report that motivates the problem, surveys related work, describes your solution, and presents an evaluation with comparisons where appropriate; plus your code with a README (and any datasets, demos, or visualizations). This is the primary deliverable — leave time to write and proofread carefully.

Modern AI tools can help improve research and engineering productivity if properly used, and this course encourages you to use them responsibly. Reasonable uses include: getting oriented in an unfamiliar paper or asking for background on a term or technique you haven't seen; talking a paper through with a model — explaining its argument back to it, asking it to push back on your reading, or having it quiz you on what you just read; brainstorming project ideas, baselines, or experiments to run; help with boilerplate, debugging, plotting, and data wrangling in your project code; and improving the clarity, grammar, and structure of your writing. There is no need to disclose or document which tools you used.

Two caveats. First, you are responsible for everything you submit, whether or not an AI wrote it. These tools confidently produce claims that are wrong, citations that do not exist, and code that silently does the wrong thing; if it appears in your proposal, report, code, or discussion, you own it, and "the model said so" is not a defense. Verify against the actual papers and your actual results. Second, you are responsible for understanding the material. Outsourcing the reading defeats the point of a seminar, and the in-class check-ins are deliberately closed-book with no electronics — the only preparation that works there is having genuinely read and thought about the papers yourself. Use AI to understand the reading faster, not to skip it.

📝 = in-class reading check-in