TypeScript Sandbox for OpenAI Vector Embeddings and AI Applications

OpenAI Vector Embeddings

Short Description

In this project, I've prepared a TypeScript-based working repository where I experimented with OpenAI, embedding logic, and vector search through various small examples.

Problem

AI topics like chat completion, in-document Q&A, image similarity, and function calling are not easily understood with a single example. I wanted to examine these topics more clearly by dividing them into small experiments within separate folders.

Solution

I divided the repository into folders such as intro, embeddings, pinecone, image-search, realtime-data, and ai-pdf-chat. In one section, I set up a simple OpenAI chat completion call. In others, I parsed PDF text, created an index with LlamaIndex, and asked questions. Additionally, I integrated streams for storing vectors with Pinecone, generating CLIP-based image embeddings, and returning product recommendations via function calling, consolidating different AI use cases into a single repository.

Result

The outcome is not a single product but rather a practical sandbox repository that explores a few fundamental ideas in AI. I particularly made the embedding, retrieval, and tool calling logic more readable by separating them into small examples.

Technologies

TypeScript, Node.js, OpenAI, LlamaIndex, Pinecone, Transformers.js, CLIP, LangChain, dotenv, PDF Parse

Setup