A take-home assignment from an internship application in May 2025. You ask a question about a stock in plain English, such as "What's happening with Palantir stock recently?", and get an answer built from current prices and news.
Gemini 2.0 Flash, through the google-generativeai SDK, decides which of five
Python functions to call: find the ticker for a company name, fetch recent
news, fetch the current price, compute the price change over a time period,
and explain the movement using the news. The repository calls these five
"agents", but they are function-calling tools that one model chooses between,
not separate agents talking to each other.
Data comes from the Alpha Vantage free tier, which allows 5 calls a minute and 500 a day, so each function waits 60 seconds and retries when it hits the limit. There is a Streamlit app, where you enter your own API keys, and a CLI.