Conversation Engineering

3.0 credits · Graduate

Conversational agents have become a primary interface to computing, but the practice of designing their behavior has not kept pace with their deployment. This course covers the specification, implementation, and evaluation of conversational agents. Topics include the structure of human conversation (turn-taking, grounding, repair, and cultural and linguistic variation); archetypes of conversation from clinical intake to customer support, each with its own goals and success criteria; the lineage of dialog systems; the system prompt as an executable behavioral specification; guardrails and the duties that attach to agent conduct; runtime concerns including retrieval, tool use, context management, and voice; and evaluation through simulated users, LLM as a judge, benchmarks, and audits of deployed systems. In a semester-long project, students specify, build, and evaluate an agent for a chosen conversation type.