Generative systems / Knowledge book
Generative AI
Overview
Models that generate. Systems that ground and evaluate.
Generation made AI impressive. Grounding makes it useful.
Grounding still needs evaluation: relevant evidence can be retrieved and used incorrectly.
Published in this topic
01When a Document Assistant Answers Without Evidence
Design a document assistant around authorization, source versions, evidence sufficiency, and citation checks rather than confidence alone.
RAG gives a model evidence. Not certainty.
Separate retrieval from generation to understand where grounded answers succeed—and where they still fail.
This collection grows section by section. Concepts describe the intended coverage; related reading may introduce only part of a section.
Start Here
Follow the numbered sections in order, or choose the subject you need. Published articles are linked where they support the section.
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- What is Generative AI?Guide planned
- Generative vs predictive AIGuide planned
- Foundation modelsGuide planned
Large Language Models
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- TokensGuide planned
- Next-token predictionGuide planned
- PretrainingGuide planned
- InferenceGuide planned
- Context windowsGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Prompting
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- Prompt basicsGuide planned
- Few-shot promptingGuide planned
- Structured promptingGuide planned
- System instructionsGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Context Engineering
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- Context selectionThe context is a design decision.
- Context limitsThe context is a design decision.
- Context compressionGuide planned
- Context qualityThe context is a design decision.
Embeddings
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- Embedding modelsGuide planned
- Vector similarityBefore generation, there was representation.
- Semantic representationBefore generation, there was representation.
RAG
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- RetrievalRAG gives a model evidence. Not certainty.
- ChunkingRAG gives a model evidence. Not certainty.
- EmbeddingsRAG gives a model evidence. Not certainty.
- Vector searchRAG gives a model evidence. Not certainty.
- GenerationRAG gives a model evidence. Not certainty.
- CitationsRAG gives a model evidence. Not certainty.
Advanced RAG
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- Hybrid searchGuide planned
- RerankingGuide planned
- Query rewritingGuide planned
- GraphRAGGuide planned
- Agentic RAGGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Fine-Tuning
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- SFTGuide planned
- LoRAGuide planned
- QLoRAGuide planned
- PEFTGuide planned
- When to fine-tuneGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Reasoning
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- Reasoning modelsGuide planned
- Test-time computeGuide planned
- Structured reasoningGuide planned
- VerificationGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Evaluation
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- GroundednessRAG gives a model evidence. Not certainty.
- RelevanceGuide planned
- FaithfulnessGuide planned
- HallucinationRAG gives a model evidence. Not certainty.
- LLM-as-judgeGuide planned
- Human evaluationGuide planned
Safety
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- GuardrailsGuide planned
- Prompt injectionGuide planned
- PIIGuide planned
- Content safetyGuide planned
Related reading
No published reading is assigned to this section yet. Its place in the learning sequence is reserved as the collection grows.
Production GenAI
Concepts in this section
Follow the related reading where available. Dedicated guides for the remaining concepts are planned.
- LatencyGuide planned
- CostGuide planned
- CachingGuide planned
- ObservabilityGuide planned
- Model routingGuide planned