Automated Tool For Generating And Providing Housing-Related Information

A computer-implemented method comprising:

• generating, by one or more computing devices and for each of multiple documents that includes housing-related information, an encoded representation for that document of the housing-related information in that document;
• generating, by the one or more computing devices while performing automated operations without human control, and for multiple initial queries related to one or more housing topics, multiple associated initial responses that are each generated by a trained large language model to a respective one of the multiple initial queries and that are each associated with one or more documents of the multiple documents that are used in proving that initial response;
• generating, by the one or more computing devices performing the automated operations without human control, feedback to use with the trained large language model based on the multiple associated initial responses, including determining, for each of the multiple initial queries, a mapping between that prior initial query and the one or more documents of the multiple documents that were used in providing the associated initial response to that initial query;
• and
• providing, by the one or more computing devices performing the automated operations without human control, and after receiving first and second queries, housing-related responses to the first and second queries that are based at least in part on the multiple documents and the trained large language model and the generated feedback, including:
• determining, by the one or more computing devices and after receiving the first query that is about at least one first housing topic, a first housing-related response to the first query based at least in part on the multiple initial queries, including:
• identifying, by the one or more computing devices, one or more initial queries of the multiple initial queries that match the first query based on one or more first defined matching criteria;
• generating, by the one or more computing devices, a first prompt to supply to the trained large language model that is based in part on the generated feedback, wherein the first prompt includes the first query and further includes one or more predetermined query-response examples and further includes information about the housing-related information of the one or more associated documents mapped to each of the one or more initial queries;
• and
• generating, by the one or more computing devices, the first housing-related response based at least in part on output of the trained large language model to the generated first prompt;
• presenting, by the one or more computing devices, the determined first housing-related response to the first query including an indication of at least one of the matching documents for the one or more initial queries as a source for the first housing-related response;
• determining, by the one or more computing devices and after receiving the second query that is about at least one second housing topic, a second housing-related response to the second query using the trained large language model, including:
• determining, by the one or more computing devices, at least one of the multiple documents whose encoded representation matches an encoded version of the second query based on one or more second defined matching criteria;
• generating, by the one or more computing devices, a second prompt to supply to the trained large language model that includes the second query and further includes the one or more predetermined query-response examples and further includes information about the housing-related information of the at least one document;
• and
• generating, by the one or more computing devices, the second housing-related response based at least in part on output of the trained large language model to the generated second prompt;
• and
• presenting, by the one or more computing devices, the determined second housing-related response to the second query including an indication of the at least one document as a source for the second housing-related response.

A non-transitory computer-readable medium having stored contents that cause one or more computing devices to perform automated operations without human input, the automated operations including at least:

• generating, by the one or more computing devices during the performing of the automated operations without human input, and for multiple initial queries from a plurality of users related to at least one housing-related topic, multiple associated initial responses that are each generated by a trained large language model to a respective one of the multiple initial queries and that are each associated with one or more documents used in proving that initial response, the one or more documents being selected from a group of multiple documents having contents including housing-related information;
• generating, by the one or more computing devices during the performing of the automated operations without human input, feedback to use with the trained large language model based on the multiple associated initial responses, including determining, for each of the multiple initial queries, a mapping between that initial query and the one or more documents of the multiple documents that were used in providing the associated initial response to that initial query;
• and
• responding, by the one or more computing devices during the performing of the automated operations without human input and using the trained large language model, to one or more queries about one or more housing-related topics, including, for each of the one or more queries:
• comparing, by the one or more computing devices, the query to the multiple initial queries to determine whether any of the initial queries match the query based on one or more defined matching criteria;
• selecting, by the one or more computing devices, one or more documents of the multiple documents to use in responding to the query, including, if one of the multiple initial queries is determined to match the query, using the at least one document associated with that one initial query as the selected one or more documents, and otherwise determining the one or more documents by matching an encoded representation of the query to encoded representations of contents of the multiple documents;
• generating, by the one or more computing devices, a prompt to supply to the trained large language model that includes the query and further includes one or more predetermined query-response examples and further includes information about the selected one or more documents;
• generating, by the one or more computing devices, a response to the query based at least in part on output of the trained large language model to the generated prompt;
• and
• providing, by the one or more computing devices, the generated response to the query including an indication of at least one of the selected one or more documents as a source for the generated response.

A system comprising:

• one or more hardware processors of one or more computing devices;
• and
• one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause at least one computing device of the one or more computing devices to perform automated operations without human control including at least:
• generating, for each of multiple documents that includes housing-related information, an encoded representation for that document of the housing-related information in that document;
• generating, for multiple initial queries related to one or more housing-related topics, multiple associated initial responses that are each generated by a trained large language model to a respective one of the multiple initial queries and that are each associated with one or more documents of the multiple documents that are used in proving that initial response;
• generating feedback to use with the trained large language model based on the multiple associated initial responses, including determining, for each of the multiple initial queries, a mapping between that initial query and the one or more documents of the multiple documents that were used in providing the associated initial response to that initial query;
• and
• providing housing-related responses that are based at least in part on the multiple documents and a trained large language model, including determining, after receiving a query about at least one housing topic, a housing-related response to the query using the trained large language model, including:
• determining at least one of the multiple documents whose encoded representation matches an encoded version of the query based on one or more defined matching criteria;
• generating a prompt to supply to the trained large language model that is based in part on the generated feedback, wherein the prompt includes the query and further includes one or more predetermined query-response examples and further includes information about the housing-related information of the at least one document;
• generating the housing-related response based at least in part on output of the trained large language model to the generated prompt;
• and
• presenting the determined housing-related response to the query including an indication of the at least one document as a source for the housing-related response.