Financial Data Analysis AI Agent

I want to stand up an autonomous AI agent that can pull raw financial data from my existing sources, clean and model it, then compile clear, executive-ready reports without additional prompting from me. The emphasis is data analysis, but the core deliverable is hands-off reporting: daily, weekly, and ad-hoc summaries that flag anomalies, spotlight KPIs, and explain the numbers in plain language.

Here’s how I picture the workflow:

1. The agent connects to files, databases, or APIs (CSV exports, cloud accounting tools such as QuickBooks or Xero, even a SQL warehouse), retrieves new records on a schedule, and performs all preprocessing in Python with Pandas/NumPy.
2. It runs the relevant calculations—ratios, period-over-period changes, variance checks—and stores intermediate results so later components can reuse them.
3. Using an LLM layer (GPT-4, Claude, etc.) through LangChain or similar orchestration, it drafts narrative insights, embeds charts created in Matplotlib or Plotly, and formats everything into PDF and HTML.
4. Finally, it pushes the finished report to my email and Slack channels and archives source logs for auditability.

I’m open to the specific stack, but prefer mainstream, well-supported tools: Python, LangChain, CrewAI or Autogen for the agentic loop, Postgres or SQLite for persistence, and standard visualization libraries. The code should be modular, readable, and accompanied by setup docs plus a one-click script or Dockerfile so I can spin it up again later.

Acceptance criteria
• End-to-end run finishes without manual intervention and produces a PDF & HTML report.
• New data dropped into the source folder triggers an updated report automatically.
• Narrative section accurately reflects the numeric findings and flags outliers.
• All credentials and API keys are handled by environment variables, not hard-coded.

If you’ve built similar agentic workflows—especially around financial data—let’s get this rolling.