embedding.EmbeddingHandler

embedding.EmbeddingHandler(self, embedding_model_name=config['llm']['embedding_model_name'], db_dir=config['llm']['db_dir'], k_matches=20)

Handles embedding operations for the Chroma vector store.

Parameters

Name	Type	Description	Default
embedding_model_name	str	The name of the embedding model to use. Defaults to the value specified in the configuration file.	config['llm']['embedding_model_name']
db_dir	str	The directory where the vector store database is located. Defaults to the value specified in the configuration file. If None then the embedding db will be non-persistent.	config['llm']['db_dir']
k_matches	int	The number of nearest matches to retrieve. Defaults to 20.	20

Methods

Name	Description
embed_index	Embeds the index entries into the vector store.
search_index	Returns k document chunks with the highest relevance to the query.
search_index_multi	Returns k document chunks with the highest relevance to the query.

embed_index

embedding.EmbeddingHandler.embed_index(from_empty=True, sic=None, file_object=None)

Embeds the index entries into the vector store.

Parameters

Name	Type	Description	Default
from_empty	bool	Whether to drop the current vector store content and start fresh.	True
sic	SIC	The SIC hierarchy object. If None, the hierarchy is loaded from files specified in the config.	None
file_object	StringIO object	The index file as StringIO object. If provided, the file will be read by line and embedded. Each line has expected format of code: description	None

search_index

embedding.EmbeddingHandler.search_index(query, return_dicts=True)

Returns k document chunks with the highest relevance to the query.

Parameters

Name	Type	Description	Default
query	str	Question for which most relevant index entries will be returned.	required
return_dicts	bool	If True, data returned as list of dictionaries, otherwise as document tuples. Defaults to True.	True

Returns

Type	Description
list[dict]	List[dict]: List of top k index entries by relevance.

search_index_multi

embedding.EmbeddingHandler.search_index_multi(query)

Returns k document chunks with the highest relevance to the query.

Parameters

Name	Type	Description	Default
query	list[str]	List of query fields (in priority order) for which most relevant index entries will be returned. e.g [industry_descr, job_title, job_descr]	required

Returns

Type	Description
list[dict]	List[dict]: List of top k index entries by relevance.