A system that can reason about a goal, select and use tools, maintain state across interactions, and operate within defined boundaries. At SAEL, an agent necessarily includes: reasoning, tool use, memory, context management, planning, execution, permissions, human-in-the-loop, and failure handling.
Dense vector representations of text (or images, audio) that capture semantic meaning. Similar concepts have similar vectors. Used for retrieval, clustering, classification, and similarity search. At SAEL, we use text-embedding-3-large (OpenAI), BGE-M3, and Nomic-Embed-Text depending on the task.
A neural network trained on vast text corpora to predict and generate human-like language. Modern LLMs (GPT-4o, Claude 3.5, Gemini, Llama 3) exhibit reasoning, coding, and tool-use capabilities. At SAEL, we treat LLMs as interchangeable reasoning engines selected per task via a model router.
A technique that combines information retrieval with language generation. Instead of relying solely on the model's training data, RAG retrieves relevant documents from a knowledge base and provides them as context to the model, enabling grounded, citable, and up-to-date responses.
A capability where an LLM can invoke predefined functions with structured arguments. The model outputs a function name and JSON arguments; the system executes the function and returns the result to the model. Enables agents to interact with APIs, databases, and external systems.