> For the complete documentation index, see [llms.txt](https://docs.livecaller.io/livecaller/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.livecaller.io/livecaller/services/ai-assistant/knowledge-base.md). # Knowledge Base **Create the Knowledge Base for Your Assistant:** * By uploading training data in **JSON format**, or * By manually adding **key/value pairs**
## What is a vector database? This is a special type of database that stores information in the form of **lists of numbers (vectors)**. Each object—whether it’s a word, image, text, or something else—is converted into an array of numbers. In our example, the image above shows **300 dimensions**, which means each object is represented by 300 numbers:\ `[0.34, 2.35, 8.34, ...]` – this is a vector of 300 numbers. **Semantic (meaning-based) similarity**\ This represents the similarity in meaning between objects in the database. **Left side (similar objects):** * Wolf, Dog, Cat – these are close to each other because they are all animals. * Chicken – stands separately but is still nearby, as it has a meaningful connection between animal and bird. * Kitten – is closest to Cat in meaning, so if a query is related to the word “kitten,” the database will search for all words and sentences semantically related to animals, names, or other meaning-bearing concepts and return the best response. **Right side (technology companies):** * 🍌 Banana * 🍎 Apple – Apple the company * G – Google * Apple – similar to Apple Inc. ## How does it work? 1. **Embedding:** The AI model converts words/images into numbers. 2. **Placement in Space:** Objects with similar meanings are placed close to each other. 3. **Search:** When you search for something, the system finds the nearest vectors. ## What is RAG? **RAG (Retrieval-Augmented Generation)** is a technique that combines search and AI generation. It is a way for AI to answer questions based on your specific knowledge base. **The 3 main steps of RAG:** 1️⃣ **Retrieval (Search)** * The user asks a question * The system converts the question into a vector * It searches the vector database for the most relevant information * Finds the 3–5 most similar documents/fragments 2️⃣ **Augmentation** * The retrieved documents are added to the question as context * The AI receives both your question and the relevant information 3️⃣ **Generation** * The AI model analyzes the context * Generates an answer based on the specific documents * The answer is accurate and well-founded **Practical Example:** **Traditional AI:** * Question: “What is our company’s vacation policy?” * Answer: “I don’t know your specific company policy...” **With RAG:** * Question: “What is our company’s vacation policy?” * Retrieval: The system finds relevant sections about vacation in the company’s HR documents * Augmentation: This information is provided to the AI * Generation: “According to your company policy, employees have 20 days of paid vacation per year, plus 10 additional days for medical leave…” {% hint style="info" %} **Question → \[Vectorization] → Vector Search → Relevant Documents**\ ↓\ **AI Model ← \[Context] ←**\ ↓\ **Answer** {% endhint %} **Use Cases:**\ 📚 **Documentation** – Company internal knowledge base\ 🏥 **Healthcare** – Analysis of medical records\ ⚖️ **Legal** – Search through legal documents\ 🛍️ **E-commerce** – Intelligent product recommendations\ 📞 **Customer Service** – Automated support