📖 Glossary
Quick, jargon-light definitions for the vocabulary used across the projects and notes. Filter by keyword or category.
22 terms
- LLM (Large Language Model)Models & generation
- A neural network trained to predict the next token, producing human-like text.
- TokenModels & generation
- A chunk of text (word piece) the model reads and generates one at a time.
- Context windowModels & generation
- The maximum number of tokens a model can consider at once — its working memory.
- TemperatureModels & generation
- A sampling knob: low = focused/deterministic, high = creative/varied output.
- System promptCore concepts
- Instructions that set persona, rules and tone for the whole conversation.
- Prompt engineeringCore concepts
- Designing inputs (instructions, examples, context) to steer model behaviour.
- Few-shot promptingCore concepts
- Giving the model a handful of examples in the prompt to demonstrate the task.
- GroundingCore concepts
- Answering from supplied source material instead of the model’s memory, to cut hallucination.
- HallucinationCore concepts
- A confident but false or unsupported statement from the model.
- EmbeddingRetrieval & RAG
- A vector representation of text where similar meanings sit close together.
- Cosine similarityRetrieval & RAG
- A score for how aligned two vectors are — used to rank documents by relevance.
- Vector databaseRetrieval & RAG
- A store that indexes embeddings for fast nearest-neighbour (similarity) search.
- ChunkingRetrieval & RAG
- Splitting documents into passages so retrieval can return focused, relevant pieces.
- Top-k retrievalRetrieval & RAG
- Keeping the k most similar chunks to the query as candidate context.
- RAGRetrieval & RAG
- Retrieval-Augmented Generation: retrieve relevant chunks, then generate a grounded answer.
- Re-rankerRetrieval & RAG
- A second-stage model that re-orders retrieved chunks by true relevance.
- HyDERetrieval & RAG
- Hypothetical Document Embeddings: embed a drafted answer to improve retrieval.
- AgentAgents
- An LLM in a loop that can use tools, observe results, and decide the next step.
- Tool / function callingAgents
- The model emitting a structured request to run external code or APIs.
- Agentic RAGAgents
- RAG wrapped in a reasoning loop that judges relevance and retries retrieval.
- Multi-agent systemAgents
- Several specialised agents (e.g. researcher, writer) that hand off to each other.
- OrchestratorAgents
- A manager agent that plans a task and delegates sub-tasks to workers.