[FEATURE] Auto tuning of knn field configuration

[heemin32]

Is your feature request related to a problem?

For the knn field type, customers can choose from a variety of parameters such as engine, algorithm, quantization technique, and more (see documentation
). Finding the optimal values for a given dataset can be time-consuming, so it may be helpful if the search relevance tool could assist users in selecting the right parameters more easily.

What solution would you like?

Users ingest vector embeddings into an index. If ground truth is not provided, the search relevance tool calculates it using an exact k-NN search. The tool then explores various parameter combinations and generates a report detailing recall, latency, and memory usage, enabling users to select the best configuration for their dataset.

What alternatives have you considered?

User do the benchmark test for each configuration by themselves.

Do you have any additional context?

N/A