Things to remember but not to implement right away. Wanna-have, nice-to-have, look-into

Year. Extract year from the chunk, store it into metadata, then add filter when retrieving, if query too has a year.
Rankers. Should we use rerankers in our pipeline? What does it mean. Get from the vector storage a lot more results, than needed. Rerank them with reranker model, then feed the limited rest further into our pipeline.