# ROADMAP

## Investigation & Exploration

* QA specific embedding models? 
* instruction tuned embeddings?
* bredth vs depth = separate workflows
* common model attributes - temp & top-k
* Evaluation metrics, how good is it doing? 
    rate my response!? 

## Planned Next

* database retrieve for tag or entity

## Planned Later

* entity chunking & re-ranking
* Logging in Ingestion
* More robust ingestion - llm response sometimes out of expected


## Done

* AI in the middle - make the llm generate multiple queries for a wider search