AI Product Development

What separates a product from a project, and the principles of responsible computing that anchor the rest of the semester. Students leave with a shared vocabulary for the course: stakeholders, problem statements, constraints, and the difference between a feature, a product, and a system.

A working knowledge of large language models — what they can and cannot do reliably — and the concrete tooling students will use across the semester (APIs, prompting patterns, evaluation, and basic deployment scaffolding). Sets the technical floor so subsequent weeks can focus on product decisions rather than tooling questions.

Customer-discovery fundamentals: how to design and run interviews that surface real problems instead of confirming your hypothesis. Covers question framing, listening discipline, mining for pain points, and translating user language into product requirements. Students prepare for their first round of conversations with newsroom and civil-society partners.

Students form project teams around problem areas in journalism, fact-checking, digital literacy, and healthcare, and are matched with partner organizations. Short in-class pitch-practice exercises establish the rhythm of presenting that the course returns to throughout the semester.

Teams sharpen the problem they are solving — tightening scope, naming the user, naming the moment of pain — and run their first round of structured customer interviews. Coursework focuses on stress-testing assumptions before writing any code.

Teams pitch their initial product ideas to the class: the user, the problem, the proposed solution, and what they have already learned from interviews. A first real test of presenting and defending an early concept, with peer and instructor feedback that feeds directly into the next iteration.

One-on-one sessions with each team to work through technical decisions, unblock specific challenges, and brainstorm approaches together. Teams keep testing prototypes with users and refining the problem statement based on what they learn.

Designing and integrating AI agents into product workflows: when an agent is the right pattern, when it is not, how to structure tool use, and how to ground outputs in retrieved context. Equally important — what a solid engineering backbone looks like underneath an AI feature so the product is reliable, testable, and explainable.

Sayli (Georgia Tech alumna) on designing AI tools that real users can actually pick up and use, drawn from her work on AI products for finance professionals. Covers the foundations of UI/UX for AI-assisted workflows — legibility, trust, error handling, and the small interaction details that decide whether a powerful model gets adopted at all.

Savannah on the parts of building a product that engineers often underweight: PR, branding, and marketing for early-career builders. How to talk about what you have made, how to position it in the right room, and how to translate technical work into language partners and funders understand.

Teams sit down with their newsroom and civil-society partners with a working prototype, gather structured feedback, and iterate on the product. The emphasis is on changing what users actually said is broken — not what the team finds most interesting to build.

Slides finalized, prototypes locked, narratives tightened. Dry-runs in class with structured feedback rounds: what is the one sentence a partner should walk away remembering, and does every slide earn its place toward that sentence?

Teams pitch finished prototypes to the room with Marc Ratkovic (University of Mannheim) and Sebastian Kotthöver (Deutsche Welle) present. The audience asks the kinds of questions a real partner would — about reliability, scope, cost, ethics, and what would have to be true for adoption.

Each team submits a finished product and accompanying report to the partner organization, documenting what was built, what was learned, the limitations of the current system, and a recommended path forward. The deliverable is built to be useful to the partner after the course ends — not just to be graded.