kaniikaaaa/financerag

FinanceRAG is a tool that answers questions about recent stock market news by pulling in real financial headlines and citing its sources. Instead of relying on a language model's built in knowledge, which can be outdated or made up, this project builds its own small search engine over fresh news and only answers using what it finds there. On a schedule, the system pulls headlines from Yahoo Finance and converts each one into a numerical representation called an embedding, using an OpenAI embedding model. These are stored in PostgreSQL with a plugin called pgvector, which lets the database search by meaning instead of by exact words. When someone asks a question like what is driving NVDA today, the system finds the most relevant recent headlines and blends how closely they match the question with how recently they were published, so a question about today's news actually returns today's news rather than something from last month. Every answer comes with its sources: the article title, publish time, and ticker symbol. If the system cannot find supporting articles, it refuses to answer rather than making something up. Because the news is pre processed and stored ahead of time, answering a question only takes one database lookup and one call to GPT-4o-mini, so responses come back quickly. The project is built with FastAPI for the web server, PostgreSQL with pgvector for storage and search, SQLAlchemy for database access, and yfinance for pulling the news data. Setting it up means cloning the repository, installing Python dependencies, adding an OpenAI API key and a database connection string, then running a couple of setup scripts before starting the server. It exposes a single main endpoint that accepts a question and returns an answer along with its cited sources. It is released under the MIT license, which allows free use including commercial use.