L02 · Structured Outputs & Tool Calling
3-hour agentic baseline: schema-constrained JSON, structured outputs, and safe tool calling for downstream APIs — with breaks.
Loading…
Course series
Twelve three-hour lectures from foundational LLM systems and Agile AI delivery through production deployment and executive handover — bilingual catalog entries for en and zh-CN.
22 courses in this series
Lecture accents
Sign in to create a classroom for this series and load lectures one at a time.
Sign in3-hour agentic baseline: schema-constrained JSON, structured outputs, and safe tool calling for downstream APIs — with breaks.
3 小时 Agentic 基线:schema 约束 JSON、结构化输出与安全工具调用(含休息)。
3-hour reasoning lab: Reflection, ReAct, and Plan-and-Solve loops for task decomposition and self-correction — with breaks.
3 小时推理实验:Reflection、ReAct 与 Plan-and-Solve 任务分解与自我纠错(含休息)。
3-hour graph lab: nodes, edges, conditional routing, and deterministic state machines that prevent runaway agents — with breaks.
3 小时图工作流:节点、边、条件路由与防止失控的确定性状态机(含休息)。
3-hour multi-agent lab: specialist roles, shared state, protocols, and supervisor orchestration — with breaks.
3 小时多智能体:专长角色、共享状态、通信协议与 supervisor 编排(含休息)。
3-hour MCP & Skills lab: Host/Client/Server, tools/resources/prompts, and Agent Skills (SKILL.md) — with breaks.
3 小时 MCP 与 Skills:Host/Client/Server、工具/资源/提示与 Agent Skills(含休息)。
3-hour vibe-coding lab: coding agents ship faster while humans review and approve every change under HITL — with breaks.
3 小时氛围编程:编码智能体加速交付,人类在 HITL 下审查并批准每一处变更(含休息)。
3-hour security lab: red-team workflows, prompt injection, exfiltration, tool hijacking, and privilege escalation — with breaks.
3 小时安全实验:红队流程、提示注入、外泄、工具劫持与权限提升(含休息)。
3-hour production lab: multi-layer memory, durable state backends, and eval pipelines for accuracy, drift, and performance — with breaks.
3 小时生产工程:多层记忆、持久状态后端与准确率/漂移/性能评测流水线(含休息)。
3-hour governance lab: runtime guardrails, cost/scope ceilings, and HK PCPD-aware controls — with breaks.
3 小时治理实验:运行时护栏、成本/范围上限与香港 PCPD 感知控制(含休息)。
3-hour production gate: deploy, harden, and evaluate a live AI system against agreed performance, safety, and reliability criteria — with breaks.
3 小时生产关卡:部署、加固并评测线上 AI 系统,对照约定的性能、安全与可靠性标准(含休息)。
3-hour decision gate: executive panel defence, governance approval, and operational handover with a complete documentation pack — with breaks.
3 小时决策关卡:高管答辩、治理批准与带完整文档包的运营移交(含休息)。