A multi-agent system that researches companies, improves its investment thesis with an analyst-critic loop, and writes bull and bear reports in parallel. It runs on GKE Autopilot.
Give it a company name or a stock ticker, and it returns an investment report with a bull case and a bear case.
A researcher gathers background on the company. An analyst turns it into an investment thesis, and a critic pushes back. They go around again until the critic accepts the thesis or three rounds have passed. The limit keeps the loop from running forever. Then two writers take the final thesis and write the bull and bear cases at the same time.
In ADK terms, the root is a SequentialAgent and the research team is a LoopAgent. The thesis moves between agents through shared state: PROMPT → research → INVESTMENT_THESIS.
Built with Google ADK 1.27.5, Gemini 2.5 Flash, LangChain, FastAPI and Uvicorn, running on GKE Autopilot.