Chapter 15: Generative AI Infrastructure & Large Scale LLM Orchestration

Generative AI at scale: vector databases, retrieval-augmented generation, distributed training, LLM serving, AI agents, fine-tuning, and model alignment.

• Vector Databases & Billion-Scale Retrieval: Pinecone, Qdrant, Milvus, Similarity Metrics (Cosine, L2, Dot Product), Approximate Nearest Neighbors (HNSW, IVF-PQ), ScaNN, and DiskANN Out-of-Core Vector Search
• Retrieval-Augmented Generation (RAG): Chunking Frameworks, Hybrid Search, Dense/Sparse Embeddings, and Re-ranking
• Distributed Model Training: Data Parallelism, Tensor Parallelism, Pipeline Parallelism (DeepSpeed, Megatron-LM)
• High-Throughput LLM Serving Frameworks: vLLM, PagedAttention, KV Caching, Continuous Batching, Speculative Decoding, and Prompt Caching
• AI Agent Systems: Tool-Calling Mechanics, Long/Short-Term Memory Stores, Reasoning Frameworks (ReAct), and Multi-Agent Orchestration
• Fine-Tuning & Model Alignment: Parameter-Efficient Fine-Tuning (PEFT, LoRA, QLoRA), Reinforcement Learning Alignment (RLHF, DPO)
• Chapter 15 References