Generative AI Learning Track & Deep-Dive Guides

Welcome to the deep-dive engineering guides for the Master AI Architecture Program.

While the Curriculum Plan serves as the source of trust and roadmap, these learning modules are written for Senior Software Architects and Engineers transitioning into AI Systems Engineering. Concepts are framed through classical systems engineering principles: LLMs as stochastic runtime engines, Pydantic schemas as domain value objects, LangFuse as OpenTelemetry distributed tracing, and LangGraph as durable state machines.

---

Active Learning Modules

Phase 1: The Engine & Systematic Prompt Architecture

Deconstructing LLMs as non-deterministic runtime engines and mastering token-level generation controls, deterministic output formatting, and prompt defense mechanisms.

1. 1.1 Spiky Intelligence & Engine Knobs: Top-$P$, Temperature, Top-$K$, Min-$P$, Presence/Frequency penalties, and token sampling mathematics.
2. 1.2 Systematic Prompt Architecture: Context delimitation, few-shot demonstration distributions, system prompt anchoring, and Pydantic v2 domain contracts.
3. 1.3 Observability & LangFuse Tracing: OpenTelemetry semantic conventions, parent-child span trees, latency profiling, and token cost attribution.
4. 1.4 Security: Injections & NIST AML Defenses: Direct/Indirect prompt injection, Unicode normalization (NFKC/NFC), sandwich defenses, and dual-boundary sanitization.
5. 🧪 Phase 1 Lab: The Hardened Classifier: Lab requirements, test cases, and Rubric benchmarks.

---

Phase 2: Reasoning Engineering & Cognitive Architecture

Engineering dynamic reasoning trajectories, search algorithms across reasoning spaces, and runtime alignment verifications.

1. 2.1 Logical Decomposition: CoT vs. Chain of Draft (CoD): Chain of Thought, Chain of Draft token compression, trade-offs between latency and semantic precision.
2. 2.2 Strategic Exploration: Tree of Thoughts & Self-Consistency: Graph-based decision spaces, BFS/DFS search strategies, and temperature voting consensus.
3. 2.3 Interactive Reasoning: ReAct Runtime Architecture: Thought-Action-Observation loops, error recovery state machines, and finite iteration boundaries.
4. 2.4 Trajectory Analysis & AlignmentCheck: Hallucination detection, trajectory verification, alignment validators, and automated circuit breaking.
5. 🧪 Phase 2 Lab: The Self-Healing Diagnostic Engine: Multi-step diagnostic specification, recovery loop design, and automated test rubrics.

---

Upcoming Curriculum Phases

The remaining phases will be populated as you progress through the roadmap:

Phase	Core Domain	Architectural Focus	Plan Reference
Phase 3	Modular Chaining	LangChain Expression Language (LCEL), Streaming, Async Pipelines	Phase 3 Plan
Phase 4	Context & Agentic RAG	HyDE, Multi-Query, Parent-Document Retrieval, Rerankers, Vector Stores	Phase 4 Plan
Phase 5	Harness Engineering	Model Context Protocol (MCP), Secure Tool Execution, Dynamic Schemas	Phase 5 Plan
Phase 6	Stateful Graphs & HITL	LangGraph, Cyclic Graphs, Human-in-the-Loop, Checkpointing	Phase 6 Plan
Capstone	NEXUS Deep Researcher	Multi-Agent Orchestration, Synthesis, Report Generation	Capstone Plan
Phase 8	Production Evaluation	Evals Flywheel, RAGAS, DeepEval, Synthetic Test Generation	Eval Plan
Phase 9	Advanced Ecosystems	Small Language Models (SLMs), Distillation, Local Inference (vLLM/Ollama)	Ecosystem Plan
Phase 10	DeepAgents Rapid Track	Production Agent Frameworks, Real-World Deployments	DeepAgents Plan