Building AI Agents with Node.js

·         From LLM call to Agent

• Chat vs agent
• Tool calling
• Deterministic vs creative outputs
• Hands-on: simple chat using LangChainJS
• Hands-on: structured output (Zod)
• Hands-on: define tools (calculator, getCurrentTime)
• Hands-on: convert chat – agent that chooses tools

·         Tools done correctly (reliability & safety) Tool schemas

• Guardrails
• Failure handling
• Hands-on: tool wrapper (validation, timeouts, error shaping)
• Hands-on: add web/API tool (mock or real)
• Common failures: hallucinated parameters, infinite loops
• Mini-comparison: LangChainJS vs another framework

·         Memory + RAG

• Retrieval > “long memory”
• Grounded answers
• Hands-on: load documents (markdown/PDF)
• Hands-on: RAG pipeline (chunk, embed, retrieve)
• Hands-on: agent answers only from sources
• Hands-on: conversation summary memory

·         Real agent workflow

• Single-agent vs multi-step workflows
• Planning + execution
• When NOT to use multi-agent systems
• Hands-on: plan, execute, verify loop
• Example task: research, summarize, produce final answer
• Persist run state

·         Production mindset + framework comparison

• Testing & evaluation
• Observability
• Hands-on: logging, run IDs, token/cost tracking