Map of Contents. LLM-based agents: how they remember, how they are evaluated, and how they interact in real time. Built on Large Language Models and Reinforcement Learning.
Memory & State
- Context Graphs - Memory graphs, workflow state machines, and change graphs for agents, with a healthcare example.
Evaluation
- Agentic Information Traversal - Benchmarks and tools for measuring how agents navigate web pages and knowledge graphs.
Conversational Agents
- Full-Duplex Conversational Agents - Voice and video agents that listen and speak at once (Moshi, GPT-Live, Tavus Griffin), benchmarks, and open problems.