RAG Validation for Global Search
Overview
There are various methods for search and retrieval in different domains supported by Global Search. It is currently unclear which retrieval methods, paired with which foundational models, perform best for each feature.
There are several potential areas for validation:
- end to end evaluation in which performance for each feature with each retrieval method and each foundational models is baseline and compared
- granular assessment of retrieval methodologies
Areas for Validation
- retrieval implementations
- BM25
- Zoekt
- semantic embedding
- Ctags
- hybrid search
- chunking strategies
- extent that dimensionality of vector embeddings impacts retrieval results; quality vs cost/latency
References
https://docs.gitlab.com/ee/user/search/
Edited by Susie Bitters