Phase 5: AI Engineering (W19–W22)
Exit criteria
- Integrate LLMs in Java (Spring AI / LangChain4j) and Go: streaming, structured output, tool calling, retries, timeouts, cost tracking
- Build a RAG system with hybrid search + reranking, and measure it (retrieval recall@k, answer faithfulness)
- Build a LangGraph agent with state, checkpoints, and human-in-the-loop approval
- Write an MCP server (tools + resources) and connect it to a real client
- Explain prompt injection, and add guardrails and evals to CI
- Ship all 12 use cases in Orbit - Use Cases with an eval suite each
- Ship Orbit v4 - Agents, RAG & MCP
🔥 This phase’s work
| Track | Where |
|---|---|
| 🎯 Weekly assignments, labs, cognitive tasks, open questions | Assignments - Phase 5 (A19–A22) |
| ⚙️ Internals | LLM Inference Internals |
| 🧩 Design patterns | AI & Agent Patterns · Anti-Patterns & Code Smells |
| ⚫ Boss fight | Red-Team Day: 30+ attacks on your own agents, fixed and reported |
| 🧠 Daily/weekly drills | Cognitive Drills · self-grade with Grading Rubric |
Tracks
AI Engineering Roadmap · LLM Fundamentals · RAG · LangChain & LangGraph · Agents & Workflows · MCP · Evals, Guardrails & LLMOps · Spring AI
Reading
- ⭐ AI Engineering (Chip Huyen, 2025)
- Hands-On Large Language Models (Alammar & Grootendorst): for visual intuition
- Build a Large Language Model (From Scratch) (Raschka): optional deep dive, 1 chapter/week
- Anthropic “Building effective agents” + the prompt engineering / tool use docs
- LangChain Academy: Introduction to LangGraph (free)
- modelcontextprotocol.io: spec + quickstarts
- Hamel Husain’s and Eugene Yan’s posts on evals and LLM patterns
DSA maintenance
1.5 h/day: alternate “2 mediums” and “1 hard” days; keep doing contests.
Retro
- What stuck:
- What didn’t:
- Skill Matrix delta: