Hello everyone, I’m from Indie Dev and I’ve been working mostly on the “ugly” side of LLMs: debugging real RAG and retrieval pipelines.
After breaking production one too many times, I started to systematically classify the ways things fail. That became an open source project called WFGY:
- a 16-problem “ProblemMap” for RAG failures (from data ingestion and embeddings to vector stores and evaluation)
- and now WFGY 3.0, a TXT-based “tension reasoning engine” that describes how pressure moves inside systems, products, or even story worlds
I joined this community because I want to connect with people who care about:
- turning theory and courses into robust real-world systems
- making LLM apps that are testable and debuggable, not just demos
- sharing concrete failure stories so others can learn faster
If anyone is interested in RAG evaluation or failure taxonomies, feel free to reach out or take anything useful from my notes:
Github onestardao WFGY