pathwaycom/pathway — reverse-engineered prompt

Build me a Python live data framework like Pathway. I want users to write one simple Python pipeline that works for local tests, batch jobs, stream replays, and live streaming without changing the code. It should connect to common places like Kafka, PostgreSQL, Google Drive, SharePoint, and lots of other data sources, then let people clean, join, sort, window, and analyze the data in real time.

Please make it feel friendly for Python users, but fast enough for production, with a stronger engine underneath if needed. It should handle late or out of order data sensibly, keep state so jobs can restart after crashes, and be deployable with Docker and Kubernetes.

Also include helpers for AI pipelines, especially live RAG over changing documents, with document parsing, chunking, embeddings, LLM calls, and an in memory vector index. Add clear examples and a quick getting started guide. Look up current docs online if you need to.