Ch 14 — The AI Landscape Today
MoE internals, test-time compute, agent architectures, RAG pipelines, and MCP protocol
Under the Hood
A Modern Model Architectures
hub
Mixture of Experts (MoE)
Sparse routing, expert selection, DeepSeek/Llama 4/Mistral
arrow_downward
view_in_ar
Multimodal Architecture
Vision encoders, cross-attention, early vs late fusion
arrow_downward
B Test-Time Compute & Reasoning
psychology
Chain-of-Thought & Reasoning Tokens
How o1/o3/R1 think before answering
arrow_downward
speed
Compute-Optimal Inference
Scaling laws for inference, adaptive compute
arrow_downward
C Agent Architectures
smart_toy
ReAct & Agent Loops
Reason-Act-Observe pattern, tool calling, memory
arrow_downward
cable
MCP Protocol Internals
JSON-RPC, tools/resources/prompts, transport
arrow_downward
D RAG & Context Engineering
manage_search
RAG Pipeline Architecture
Embedding, chunking, retrieval, reranking, generation
arrow_downward
database
Vector Databases & Embeddings
ANN search, HNSW, cosine similarity, hybrid search
arrow_downward
E Inference & Deployment
memory
Inference Optimizations
KV cache, continuous batching, speculative decoding, quantization
arrow_downward
rocket_launch
The Full Picture
Connecting all 14 chapters — the complete AI stack