zeldaniela/chatbot-rag-vehiculos

This project is a chatbot that lets you ask questions about vehicle catalogs and technical documents in plain language, and get answers drawn directly from those documents. It is built in Python and uses a technique called RAG, short for Retrieval-Augmented Generation, which means the system first finds the most relevant pieces of your uploaded documents and then asks an AI model to compose an answer based only on those pieces, rather than making things up. To get started, you place plain text files containing vehicle manuals, model specifications, or maintenance guides into a folder called data. Then you call one API endpoint to index those files. The system breaks them into smaller chunks, converts each chunk into a numerical representation using OpenAI's embedding model, and stores everything in a local database called ChromaDB. After that, you can ask questions through a second API endpoint and the system retrieves the most relevant chunks, feeds them to OpenAI's GPT-4o-mini model, and returns a focused answer along with the source text it used. The project exposes a small web API built with FastAPI. There are three endpoints: one to check that the service is running, one to trigger indexing or re-indexing of your documents, and one to send a question and receive an answer. The whole setup runs inside Docker containers, so you can get it running on any machine without configuring Python manually. For development, you can also run it locally with a standard Python virtual environment. There are some clear constraints. Right now it only reads plain text files, so PDFs or web pages would need extra work to support. It requires an OpenAI API key and an internet connection for both embedding and answering. If you update your documents, you have to call the indexing endpoint again manually. The API also has no built-in authentication, which would be needed before putting it on the public internet. ChromaDB works well for smaller collections but would need to be swapped for a larger solution if you indexed millions of documents. The project is licensed under MIT, so it can be used or modified freely.