Most AI courses teach you to use models. This one examines how they
are built, evaluated, and deployed in production environments
where reliability matters. The curriculum draws from published
research on transformer architectures, reinforcement learning
from human feedback, and constitutional AI methods used in
systems like Claude.
Course focus areas
Participants work through the technical architecture of large
language models, then move into evaluation design, covering
benchmarks, red-teaming protocols, and alignment testing. A
dedicated section examines Claude AI specifically, analyzing
Anthropic's published work on constitutional AI and how it
differs from instruction-tuning approaches used elsewhere. This
is not a promotional section. The goal is comparative technical
analysis.
Practical components
Each module includes a lab component. Students build evaluation
harnesses, run structured prompt experiments, and document model
behavior across edge cases. The final project requires designing
a deployment checklist for a specified use case, graded by Priya
Oduya and Stellan Kirchbach, both of whom have industry
backgrounds in AI safety and systems engineering.
- 8 weeks, two live sessions per week
- Access to API credits for hands-on testing
- Peer review integrated into grading structure
- Reading list drawn from ArXiv papers published within the
last 18 months
This course does not guarantee job placement or specific
skill outcomes. Results depend on prior background and time
invested.
