Dalil
A semantic search engine for Russian-language religious documents that uses hybrid retrieval and reranking techniques. It does not generate answers; it only retrieves, ranks, and returns original source documents.
Timeline (last 90 days)
The project builds a search engine for Russian-language religious documents, with the dataset expanded from a few thousand records to over 10,000 documents.
The developer chose not to integrate an LLM to generate answers, only to retrieve, rank and return the original sources, because the LLM might ignore important conditions or piece together different rulings.
After reranking, a threshold was set so that an empty result set is returned when the best candidate's score is too low.
During data migration for the project, a new records_ru_v2 collection was created, keeping the old collection as a fallback.
All Facts (4)
The project builds a search engine for Russian-language religious documents, with the dataset expanded from a few thousand records to over 10,000 documents.
50%UnverifiedThe developer chose not to integrate an LLM to generate answers, only to retrieve, rank and return the original sources, because the LLM might ignore important conditions or piece together different rulings.
50%UnverifiedAfter reranking, a threshold was set so that an empty result set is returned when the best candidate's score is too low.
50%UnverifiedDuring data migration for the project, a new records_ru_v2 collection was created, keeping the old collection as a fallback.
50%