I have a set of nodes with metadata like below for each of my documents ingested into a PostgresDB with pgvector

When I query the documents using query like "What is the EBITDA for the year 2023 and 2020 for XYZ company?" I want to retrieve different documents based on the "date" metadata and "name" along with semantic search using LLM.

Currently metadata is not taken into consideration and documents returned are also from 2013 etc.

Is there a way to create a keyword based retriever like BM25Retriever based on an existing index? I have already embedded and ingested my documents into a postgres vectorstore and want to create a retriever from an already existing index.

Is there a way to do Recency filtering using FixedRecencyPostprocessor when the metadata 'date' key has values in %d-%m-%Y format and not in "%m-%d-%Y"

Is there a way to do metadata filtering based on multiple keys like below and using combination of "AND" / "OR" queries?
PS: I am using ChromaDB locally as vector index