Stańczyk is a simple RAG system built as an investigative assistant working with a selected source material. Its purpose is not to “know everything”, but to answer from a specific context: documents, case fragments, notes, and data previously stored in a database.

The system splits source material into chunks, stores them in SQL, and uses embeddings with contextual retrieval to find the fragments most relevant to the user’s question. Only then does the LLM generate an answer. This keeps the response anchored in the material rather than only in the model’s general knowledge.

The demo currently works only in Polish: Stańczyk

This is intentionally a simple RAG architecture: no unnecessary layers, no imitation of a magical expert system, and no attempt to replace human analysis. Its value lies in faster access to context, better information organization, and support for working with documents.

Stańczyk shows the practical value of RAG in situations where a user needs to work conversationally with a specific collection of data: documentation, a case file, an archive, a knowledge base, or legal, research, and operational material.

Technology: TypeScript, Next.js, MySQL, OpenAI API, PDF parsing, text chunking, embeddings, contextual retrieval