Imagine you are a business owner and you have tons of company documents on your computer. You also have meeting notes on Notion, H notes on Evernote, and messages on multiple channels like Slack and WhatsApp. However, you constantly waste time searching for information in multiple places. A few months ago, you heard about people using chatGPT as their personal assistant, but you were unsure how it could apply to your situation. You have private documents that may contain sensitive data, and you can’t just paste all of your documents in chatGPT. What if I told you there is a way to build a personal assistant that is fed with all of your company’s knowledge base?
In this video series, I will show you step by step how to build exactly that. In this specific video, I will focus on the different steps and concepts to understand the big picture.
The first concept is large language models (LLMs). LLMs are smart autocomplete models that can generate complete sentences based on the billions of data they have been trained on. However, creating your own LLM with your own documents is expensive and not as performant as existing models like GPT. Additionally, existing models have limitations on context length.
To overcome these limitations, we can use multiple concepts such as shunks, embeddings, and vector stores. Shunks are small pieces of data that we divide our documents into. Embeddings are numerical representations of text that capture semantic relationships. Vector stores are places where we store these embeddings.
To query the vector store, we transform the user’s question into an embedding and find the closest relevant embeddings. We then use an LLM to generate a well-formatted response to the user’s question.
In the next video, we will dive into the code and implement these solutions. Stay tuned!
