Month 1 Local LLM Foundations, Private Inferencing & Structured Tool Calling Weeks 1–4 • 8 Live Sessions
  • Session 1-2: Modern LLM runtime mechanics: Quantization techniques (GGUF, AWQ), context window management, and tokenization internals.
  • Session 3-4: Private air-gapped inference: Deploying and fine-tuning local models (Llama 3, Mistral, Qwen) using Ollama and vLLM on self-hosted instances.
  • Session 5-6: Deterministic outputs: JSON schema enforcement, Pydantic/Zod structured output binding, and resilient tool-calling fallback mechanisms.
  • Session 7-8 (Live Hand-Holding): Air-Gapped Tool Calling Engine Lab — Building a local, zero-leakage extraction engine with Ollama and TypeScript that parses messy enterprise contracts into strict database records.
Month 2 Model Context Protocol (MCP) Server Architecture & Tool Design Weeks 5–8 • 8 Live Sessions
  • Session 9-10: Model Context Protocol (MCP) deep-dive: Architecture, JSON-RPC 2.0 transport layers (Stdio vs SSE), and protocol handshakes.
  • Session 11-12: Engineering MCP Tools, Resources, and Prompts: Exposing internal REST endpoints, file systems, and Git repositories securely.
  • Session 13-14: Relational Database MCP Servers: Architecting real-time schema inspection, query execution guards, and read-only transaction wrappers.
  • Session 15-16 (Live Hand-Holding): Enterprise Database MCP Server Lab — Building a production-grade TypeScript MCP server connecting PostgreSQL to Claude Desktop/Cursor with dynamic schema discovery and safe parameter sanitization.
Month 3 Stateful Multi-Agent Orchestration, LangGraph & Human-in-the-Loop Weeks 9–12 • 8 Live Sessions
  • Session 17-18: ReAct framework limitations & graph-based agent architecture: Nodes, edges, conditional routing, and state persistence with LangGraph.
  • Session 19-20: Multi-Agent Collaboration: Hierarchical supervisor patterns, specialized worker sub-graphs, and inter-agent memory isolation.
  • Session 21-22: Human-in-the-Loop (HITL) checkpoints: Interrupt triggers, manual approval gates, time-travel debugging, and state rollback mechanics.
  • Session 23-24 (Live Hand-Holding): Autonomous Code-Review Agent Cluster Lab — Building a multi-agent system where a Git-fetcher, a security-linter, and an architectural-evaluator critique pull requests with human sign-off triggers.
Month 4 Production Hybrid RAG, n8n Workflow Automation & Capstone Defense Weeks 13–16 • 8 Live Sessions
  • Session 25-26: Advanced hybrid retrieval: Dense vector embeddings vs BM25 sparse search using PostgreSQL pgvector, contextual re-ranking, and Reciprocal Rank Fusion (RRF).
  • Session 27-28: Enterprise automation with n8n: Self-hosting n8n via Docker, webhooks, integrating custom LLM nodes, and orchestrating business workflows.
  • Session 29-30: Observability, evaluation & security: Tracing agent loops with LangSmith, tracking hallucination metrics, and mitigating prompt injection vulnerabilities.
  • Session 31-32 (Final Capstone & Architecture Defense): Live Autonomous Agent Deployment — Launching an end-to-end autonomous triage agent capable of reading production tickets, querying knowledge graphs via MCP, and executing automated fixes with real-time audit logging.