COURSE OUTLINE

Part 1: Advanced Model Engineering & Data (Days 1–4)

Day 1: The Modern LLM Stack

• Model Architectures: Deep dive into Mixture-of-Experts (MoE) and Transformer internals.
• The Gemini Ecosystem: Comparing Pro, Flash, and Nano for specific compute budgets.
• Context Engineering: Mastering the 2M+ token window and context caching.
• Hands-on: Benchmarking model performance and latency for long-context retrieval.

Day 2: Advanced RAG & Vector Infrastructure

• The Retrieval Pipeline: Moving from basic search to Semantic Re-ranking.
• Vector Databases at Scale: Deep dive into Vertex AI Vector Search and HNSW indexing.  
• Hybrid Search: Combining sparse (keyword) and dense (semantic) vectors.  
• Hands-on: Building a RAG system that queries millions of legal or technical documents.

Day 3: Fine-Tuning & Distillation

• The SFT Workflow: Supervised Fine-Tuning for specific task alignment.
• PEFT & LoRA: Using Low-Rank Adaptation to train models on commodity hardware.
• Model Distillation: Training a smaller, faster model (Gemma) using a larger teacher (Gemini 1.5 Pro).
• Hands-on: Fine-tuning a domain-specific model for medical or financial nomenclature.

Day 4: Multimodal Application Design

• Beyond Text: Integrating Vision (Imagen), Audio (Lyria), and Video (Veo) into a single app.
• Multimodal Embeddings: How to search through images and video using text queries.
• Hands-on: Building an "Automated Content Producer" that generates a script, background score, and video clip from a single prompt.

Part 2: Agents, Orchestration & Production (Days 5–8)

Day 5: Autonomous Agents & Tool Use

• Agentic Reasoning: Introduction to the ReAct (Reason + Act) loop.
• Function Calling: Teaching models to securely interact with internal APIs and databases.
• State Management: Handling long-term "Memory" and conversation state.
• Hands-on: Building an agent that can browse the web and update a SQL database autonomously.



Day 6: Multi-Agent Systems & Orchestration

• Orchestration Frameworks: Deep dive into LangGraph and Vertex AI Agent Builder.
• Collaborative Patterns: Hierarchical vs. Peer-to-peer agent architectures.
• Conflict Resolution: Managing disagreements between specialized agents.
• Hands-on: Architecting a "Digital Department" where a Manager Agent coordinates a Coder Agent and a QA Agent.

Day 7: LLMOps & Evaluation

• The Evaluation Flywheel: Using LLM-as-a-judge and automated toxicity scoring.
• Production Monitoring: Tracking token usage, latency, and "Semantic Drift."
• CI/CD for AI: Implementing automated unit tests for prompts and model responses.
• Hands-on: Setting up a monitoring dashboard for a production AI service.

Day 8: Security, Governance & Capstone

• Adversarial AI: Mastering Prompt Injection and Data Poisoning defenses.
• Governance: Implementing the EU AI Act compliance checks and Responsible AI filters.
• Final Capstone Implementation: Building and deploying a "Self-Correcting Enterprise Agent."
• Technical Defense: Presenting the final system's architecture and safety profile.