Phase 614 DaysAdvanced
Phase 6 — Agentic AI
Design reliable multi-step AI agents with explicit state transitions, tool use, memory, and measurable evaluation — not just prompt heuristics.
- Build LangGraph agents with deterministic state machines and tool execution.
- Evaluate agent reliability with task suites across easy, medium, and failure cases.
- Implement safety guardrails against prompt injection and unsafe tool invocation.
⚡ Must Know
- ReAct Pattern — reason + act loop
- Tool Use + Function Calling
- LangGraph — state graphs, nodes, edges
- Agent Memory — short-term, long-term, episodic
- Human-in-the-Loop + Interrupt patterns
- Multi-Agent Orchestration — CrewAI, AutoGen
- Agent Evaluation — task harnesses, pass@k
- Prompt Injection + Security
- Tracing + Observability — LangSmith
✨ Good to Know
- Semantic Kernel
- OpenAI Assistants API + Threads
- Tavily Search API integration
- Mixture-of-Agents approaches
📚 Resources
🏗️ Projects
Web Research Agent
Planning, search, and synthesis nodes returning source-backed research briefs.
Multi-Agent Code Reviewer
Specialist agents reviewing PRs for bugs, style, tests, and security.
Customer Support Agent
Support assistant with memory, policy grounding, and escalation triggers.