LLM Prompt Workflows for Token-Limited Source-to-Pay Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Source-to-Pay platforms face challenges in deploying Generative Artificial Intelligence (GenAI) functionality due to limited processing capacity and unconstrained creativity, leading to inaccurate information generation, and require unique implementation tailored to different enterprise maturity levels.
Innovation Solution
A method and platform that utilize a large language model (LLM) by dividing context data into manageable chunks, determining user intent, and selecting appropriate processing workflows to generate prompts, ensuring compatibility with diverse enterprise needs and improving user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If context data is processed in full batches, then processing completeness is improved, but processing capacity is exceeded due to token limits
Solution Approach 1:
The patent divides context data into multiple smaller batches, each containing a manageable number of tokens that fit within LLM processing limits. This segmentation allows the system to process large volumes of data by breaking them into chunks that can be handled individually, thereby maintaining processing completeness without exceeding capacity constraints.
2Ease of operation
If LLM processing is implemented, then user interaction is simplified, but accuracy deteriorates due to unconstrained creativity
Solution Approach 1:
The patent modifies the operational parameters of the LLM by setting temperature to 0 and top_p to 0.1, which constrains the model's creative variability and ensures deterministic, accurate responses. This parameter adjustment maintains simplicity of user interaction while preventing accuracy deterioration from unconstrained generation.
Solution Approach 2:
The system validates LLM responses against ground truth data and corrects inaccuracies through feedback loops. This ensures that while the LLM provides simplified user interaction, the information accuracy is maintained by comparing outputs with verified data and making necessary corrections.
3Adaptability or versatility
If custom workflows are developed for each enterprise, then adaptability is improved, but device complexity increases
Solution Approach 1:
The patent implements a universal workflow engine that can handle multiple enterprise scenarios through a single platform. The system provides pre-configured workflows that can be selected and customized without requiring complex custom development, thereby achieving enterprise adaptability while minimizing implementation complexity.
Data Source
AI summary
There is provided a computer implemented method of using a large language model (LLM) in a Source-to-Pay platform. The method comprises receiving a user query associated with context data; estimating a number of tokens associated with the context data and dividing the context data into a plurality of context data parts each having an estimated number of tokens below a maximum number of tokens per transaction associated with the LLM; determining a user intent from the user query and selecting one or more of a plurality of predetermined processing workflows based on the user intent; processing the context data parts according to the selected workflow to generate one or more LLM prompts each comprising a respective context data part; forwarding the one or more LLM prompts to the LLM and receiving a respective response; and using at least some content from the one or more responses to generate an answer to the user query.