Matrix user interface for LLM-powered data analysis and generation
A data analytics system uses a grid-based data structure to improve the usability of LLMs in the analysis of large data sets, to synthesize information for use in other generative AI contexts, and to improve a user's ability to interface with an LLM. A grid-based data structure is a data structure or database that stores the results of column prompts applied to sources. The grid-based data structure may store the results in a relational manner. For example, a grid-based data structure may have rows that correspond to sources (e.g., documents, files, or databases) and columns that correspond to prompts. Each cell of the grid-based data structure stores the output of the column prompt applied to a source using an LLM. Thus, each column prompt may be systematically applied to each source to generate information based on the sources in an organized way.
1. A method comprising:
receiving a plurality of sources at a data analytics system, where each of the plurality of sources comprises a data file;
storing a grid-based data structure at the data analytics system, where the grid-based data structure comprises a set of rows, a set of columns, and a set of cells, where each cell of the set of cells has a corresponding row of the set of rows and column of the set of columns, where each row of the set of rows has a unique corresponding source of the plurality of sources, where each column of the set of columns has a unique corresponding column prompt, and where each column prompt is free text for a prompt to apply to each of the plurality of sources to generate values for cells in the corresponding column;
displaying a user interface to a user through a client device, wherein the user interface comprises a matrix user interface that displays the set of rows, the set of columns, and the set of cells of the grid-based data structure, where each row of the set of rows is displayed with an indication of the corresponding source of the row, and where each column is displayed with an indication of the corresponding column prompt of the column;
populating each cell of the set of cells of the grid-based data structure by, for each cell of the set of cells of the grid-based data structure:
identifying a source of the plurality of sources corresponding to a row of the cell;
identifying a column prompt corresponding to a column of the cell;
generating a final prompt for the cell based on the identified source and the identified column prompt, wherein the final prompt comprises text from the identified column prompt and text from the identified source;
transmitting the final prompt to a large language model;
receiving a response to the final prompt from the large language model; and
updating the matrix user interface to display a populated version of the cell based on the received response to the final prompt, where the populated version of the cell comprises at least a portion of the received response.
2. The method of claim 1 , further comprising:
for each of the plurality of sources, generating a plurality of chunks of the source, where each chunk comprises a set of sections of the source that are semantically related to each other; and
for each cell of the set of cells of the grid-based data structure:
generating a plurality of context window prompts, wherein each context window prompt comprises the column prompt associated with the column of the cell and a subset of the plurality of chunks of the source associated with the cell;
receiving responses to the plurality of context window prompts from a large language model; and
generating the final prompt for the cell based on the received responses to the plurality of context window prompts.
3. The method of claim 2 , wherein each section of the set of sections corresponds to a distinct portion of the corresponding source based on a structure of the source.
4. The method of claim 2 , wherein each chunk comprises a set of sections that are sequential within the source.
5. The method of claim 2 , wherein generating a context window prompt comprises:
identifying a subset of the plurality of chunks with a total token size less than a threshold value.
6. The method of claim 1 , wherein receiving the plurality of sources comprises:
receiving locator information for each of the plurality of sources.
7. The method of claim 1 , further comprising:
receiving each column prompt through a prompt user interface displayed to the user by the client device.
8. The method of claim 7 , wherein receiving each column prompt comprises:
receiving parameters for prompting the large language model through the prompt user interface.
9. The method of claim 1 , further comprising:
receiving an indication from the user through the client device to collapse a subset of the set of rows; and
responsive to receiving the indication, updating the matrix user interface to display a single row for the subset of rows, wherein each cell associated with the single row displays cell contents for a single source of a subset of sources corresponding to the subset of rows.
10. The method of claim 1 , wherein generating the final prompt comprises:
generating a final prompt for multiple cells of the set of cells, wherein the final prompt comprises multiple column prompts corresponding to the multiple cells.
11. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform operations comprising:
receiving a plurality of sources at a data analytics system, where each of the plurality of sources comprises a data file;
storing a grid-based data structure at the data analytics system, where the grid-based data structure comprises a set of rows, a set of columns, and a set of cells, where each cell of the set of cells has a corresponding row of the set of rows and column of the set of columns, where each row of the set of rows has a unique corresponding source of the plurality of sources, where each column of the set of columns has a unique corresponding column prompt, and where each column prompt is free text for a prompt to apply to each of the plurality of sources to generate values for cells in the corresponding column;
displaying a user interface to a user through a client device, wherein the user interface comprises a matrix user interface that displays the set of rows, the set of columns, and the set of cells of the grid-based data structure, where each row of the set of rows is displayed with an indication of the corresponding source of the row, and where each column is displayed with an indication of the corresponding column prompt of the column;
populating each cell of the set of cells of the grid-based data structure by, for each cell of the set of cells of the grid-based data structure:
identifying a source of the plurality of sources corresponding to a row of the cell;
identifying a column prompt corresponding to a column of the cell;
generating a final prompt for the cell based on the identified source and the identified column prompt, wherein the final prompt comprises text from the identified column prompt and text from the identified source;
transmitting the final prompt to a large language model;
receiving a response to the final prompt from the large language model; and
updating the matrix user interface to display a populated version of the cell based on the received response to the final prompt, where the populated version of the cell comprises at least a portion of the received response.
12. The computer-readable medium of claim 11 , the operations further comprising:
for each of the plurality of sources, generating a plurality of chunks of the source, where each chunk comprises a set of sections of the source that are semantically related to each other; and
for each cell of the set of cells of the grid-based data structure:
generating a plurality of context window prompts, wherein each context window prompt comprises the column prompt associated with the column of the cell and a subset of the plurality of chunks of the source associated with the cell;
receiving responses to the plurality of context window prompts from a large language model; and
generating the final prompt for the cell based on the received responses to the plurality of context window prompts.
13. The computer-readable medium of claim 12 , wherein each section of the set of sections corresponds to a distinct portion of the corresponding source based on a structure of the source.
14. The computer-readable medium of claim 12 , wherein each chunk comprises a set of sections that are sequential within the source.
15. The computer-readable medium of claim 12 , wherein generating a context window prompt comprises:
identifying a subset of the plurality of chunks with a total token size less than a threshold value.
16. The computer-readable medium of claim 11 , wherein receiving the plurality of sources comprises:
receiving locator information for each of the plurality of sources.
17. The computer-readable medium of claim 11 , the operations further comprising:
receiving each column prompt through a prompt user interface displayed to the user by the client device.
18. The computer-readable medium of claim 17 , wherein receiving each column prompt comprises:
receiving parameters for prompting the large language model through the prompt user interface.
19. The computer-readable medium of claim 11 , the operations further comprising:
receiving an indication from the user through the client device to collapse a subset of the set of rows; and
responsive to receiving the indication, updating the matrix user interface to display a single row for the subset of rows, wherein each cell associated with the single row displays cell contents for a single source of a subset of sources corresponding to the subset of rows.
20. The computer-readable medium of claim 11 , wherein generating the final prompt comprises:
generating a final prompt for multiple cells of the set of cells, wherein the final prompt comprises multiple column prompts corresponding to the multiple cells.