Week 15: Multi-Agent Systems, Routing & Orchestration

Weekly Learning Objectives

1. Day 1 (day01_when_single_agent_beats_multi_agent_and_topologies): Mathematical cost/latency/error-compounding laws of Single-Agent vs. Hierarchical Supervisor, Peer-to-Peer Swarm, and Sequential Pipeline topologies.
2. Day 2 (day02_supervisor_worker_and_structured_handoff_protocols): Building a Supervisor Router with specialized sub-agents (DatabaseAgent, K8sAgent, NetworkAgent) using strict Pydantic handoff packets to prevent context telephone game loss.
3. Day 3 (day03_parallel_map_reduce_agents_and_voting_ensembles): Fan-out / Fan-in Map-Reduce log analysis across shard agents and Condorcet Majority Voting ensembles.
4. Day 4 (day04_adversarial_debate_and_evaluator_optimizer_loops): Generator-Critic (RemediationGenerator + SecurityCritic) adversarial loop that blocks unsafe infrastructure patches before execution.
5. Day 5 (day05_agentic_token_budgets_cost_caps_and_circuit_breakers): Per-request Dollar/Token budget enforcers, hop-count circuit breakers, and loop oscillation detectors.
6. Day 6 (day06_evaluating_agents_trajectory_metrics_and_tool_accuracy): Agent Trajectory Evaluation (Tool Selection F1, Argument Exact Match, Trajectory Optimality Ratio).
7. Day 7 (day07_week15_mini_project_multi_agent_incident_war_room): End-to-End Multi-Agent Incident War Room combining Supervisor routing, Map-Reduce log inspection, Critic safety verification, and token budget governance.