Skip to content

References: LLM and AI Foundations

  1. Large Language Model - Wikipedia - Comprehensive overview of LLM architecture, training objectives, parameter scale, and emergent capabilities — directly foundational for this chapter's explanation of why enterprise-specific knowledge must be injected at inference time rather than assumed from training data.

  2. Transformer (deep learning architecture) - Wikipedia - Explains the transformer architecture, multi-head attention mechanism, and positional encoding — supporting this chapter's practitioner-level treatment of how transformers process context windows and why long-context quadratic cost is bounded in practice.

  3. Attention Mechanism (machine learning) - Wikipedia - Covers the attention mechanism that enables transformers to relate any token to any other token in the context — directly supporting this chapter's explanation of why context window position matters and how the model attends to retrieved graph context.

  4. Hands-On Large Language Models - Jay Alammar, Maarten Grootendorst - O'Reilly Media - Chapters 1–6 cover transformer internals, tokenization, prompting strategies, temperature and sampling parameters, and fine-tuning — providing practical depth for all major LLM mechanics introduced in this chapter.

  5. LLM Engineer's Handbook - Paul Iusztin, Maxime Labonne - Packt - Covers prompt engineering for production systems, few-shot prompting, instruction tuning, function calling, and evaluation metrics including faithfulness scoring — directly supporting this chapter's treatment of prompt anatomy and LLM evaluation for context graph systems.

  6. Prompt Engineering - Wikipedia - Defines prompt engineering techniques including system prompts, zero-shot and few-shot prompting, and in-context learning — directly supporting this chapter's section on the control interface for context-graph-powered LLM systems.

  7. Hallucination (artificial intelligence) - Wikipedia - Covers LLM hallucination patterns, causes, and mitigation — directly relevant to this chapter's treatment of faithfulness scoring as the primary evaluation metric for grounded enterprise decision support systems.

  8. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks - Lewis et al., arXiv - The original RAG paper introducing retrieval-grounded generation — foundational reference for this chapter's function calling section showing how agents interact with context graph APIs to conduct interactive multi-hop retrievals.

  9. Token (language model) - Wikipedia - Explains tokenization, vocabulary sizes, and token count estimation — directly supporting this chapter's two practical implications of tokenization for context graph assembly: token budget management and compact serialization of graph data.

  10. AI Alignment - Wikipedia - Covers RLHF and instruction tuning as alignment techniques that produce helpful, honest, and calibrated model behavior — supporting this chapter's explanation of why modern instruction-tuned models hedge appropriately when context is insufficient, which is the desired behavior for enterprise decision support.