Azure Single Sign-On (SSO) is a feature of Microsoft's cloud-based identity and access management service, Azure Active Directory (Azure AD). It allows users to log into multiple applications and services with a single set of credentials, eliminating the need for multiple usernames and passwords. When a user logs into one application, they are automatically authenticated for other applications they have permissions to access, without needing to sign in again.

Retrieval Augmented Generation (RAG) merges large pre-trained language models with external information retrieval systems. Presented with a query, RAG first fetches relevant documents from a corpus and then uses these as context for a Transformer-based model to generate a detailed response. This combination allows the model to access external knowledge, enhancing its versatility and informativeness in answering questions.

LangChain is a state-of-the-art open-source framework, pivotal for developing applications powered by Large Language Models (LLMs). It provides a bridge, linking these LLMs, such as OpenAI's GPT series, to a diverse range of external data sources, fostering the creation of advanced natural language processing (NLP) applications. With LangChain, developers can seamlessly integrate LLMs with other components to produce applications that range from chatbots and document summarisation to code analysis. Moreover, it streamlines the process for developers, offering modules like model interaction, data connection, and 'chains' that link multiple LLMs for more complex tasks. As a result, agents can access up-to-date information, providing users with enriched and accurate interactions.

Creating a personal experience means creating a unique user history. The requires finding a balance of long memory and how much of the context window to dedicate to memory versus other tasks. We use an advanced combination of user specific details, Conversation Summary Memory and recent interactions.

A vector database is a specialised storage system designed to handle high-dimensional vectors, commonly used in machine learning and similarity search tasks. Instead of querying data based on exact matches like traditional databases, vector databases allow users to search for data points that are "closest" or most similar to a given query vector. Using advanced indexing techniques, these databases efficiently handle large-scale datasets, enabling rapid and accurate retrieval of similar items based on cosine similarity, Euclidean distance, or other distance metrics.

We have developed a Smart Text Processing which has greatly improved retrieval compared to dumb text splitters. Data curation prior to embedding is critical to retrieval performance.

Providing our Smart Agents with Tools is what takes them from being simple chatbots to Smart Agents. Tools are essentially functions that can range from generic utilities like calculators and search, to other chains, or even other agents. These tools are crucial as they enable LLMs to access and manipulate external data sources, enhance the agent's capabilities, and facilitate complex tasks that would otherwise be beyond the scope of a standalone language model.