colossus-lab/openarg_backend

OpenArg Backend is the engine behind openarg.org, a website that answers plain-language questions about Argentine government data, things like public spending, dollar rates, central bank figures, and congressional records. You ask a question in normal words, and the system figures out where to look, pulls the relevant numbers, and writes back an answer with sources and charts. Under the hood, the question passes through a series of AI agents that each handle one job. One agent reads the question and decides which of nine different government data sources to check, ranging from economic time series and central bank indicators to legislative session transcripts and public officials' financial disclosures. Several agents then gather that data at the same time, another agent turns the question into a database query when needed, and a final agent writes up the findings with citations, a confidence level, and any charts the data supports. If the first attempt does not find enough information, the system can try again with a different approach. To avoid repeating expensive work, the system checks whether a very similar question has already been answered recently and reuses that answer instead of starting from scratch. Behind the scenes, background workers continuously scrape, download, and process new government datasets so the cached information stays current. The project is built with FastAPI for the web server, PostgreSQL for storing data, Redis and Celery for background processing, and Claude and Gemini language models to understand questions and write answers. It is organized in layers that separate the web interface, the core logic, and the connections to outside data sources and AI providers, which keeps each part easier to change independently. Getting it running requires Docker along with API keys for the AI providers and cloud services it depends on. The full README is longer than what was shown.