Make LLM 'metadata filtering' processes agentic

There are two main processes you need an LLM for for the metadata filtering task:
- Extraction of metadata filtering
- Parsing of user queries

Due to small LLMs being rather inaccurate in performing these tasks, we may need to either build a more robust pipeline for retrieving relevant data from the input or we could use agentic system (Re-Act) and let the model decide whether it has done a good job in extracting all the requested data or it needs to perform further refinement...