LLM Skill Distillation for Reusable SOP Code Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems are inefficient for employees to locate and manually interact with internal UI systems to complete standard operating procedure (SOP) tasks, exacerbated by employee attrition leading to lost expertise and inefficient knowledge transfer.
Innovation Solution
A framework using large language models (LLMs) to automatically translate textual instructions into executable code, incorporating skill distillation and composition, with a skill distillation agent, composition agent, and execution agent to streamline SOP processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If employees manually interact with internal UI systems to complete SOP tasks, then they can perform tasks step-by-step, but it is inefficient and time-consuming
Solution Approach 1:
The patent replaces manual mechanical interaction with UI systems through LLM-based automated code generation. The system translates natural language task descriptions into executable code that automatically performs SOP tasks, eliminating the need for employees to manually navigate and interact with internal UI systems step-by-step.
Solution Approach 2:
The system enables self-service by allowing employees to describe tasks in natural language and having the LLM automatically generate and execute the necessary code to complete them. This eliminates the need for employees to manually locate information in documents or interact with complex UI systems.
2Loss of information
If employees manually locate information in SOP documents, then they can find specific instructions, but it is inefficient and expertise is lost with attrition
Solution Approach 1:
The system creates digital copies of SOP knowledge in the form of generated code and stored solutions. When employees need to perform tasks, the system retrieves relevant information from stored solutions rather than requiring employees to manually search through documents, thus preserving knowledge even when employees leave.
Solution Approach 2:
The LLM acts as an intermediary between employees and SOP documents. Instead of employees directly searching through extensive documentation, the LLM understands natural language queries, retrieves relevant information from stored solutions, and presents it in a usable format, thereby retaining knowledge without requiring employees to manually locate information.
3Productivity
If the system generates code from natural language prompts, then code generation is expedited, but the system complexity increases
Solution Approach 1:
The system segments the code generation process into distinct components: a skill distillation agent that breaks down complex tasks into atomic skills, a composition agent that combines skills into complete solutions, and an execution agent that runs the generated code. This segmentation manages complexity by dividing the overall system into manageable, specialized modules.
Solution Approach 2:
The system performs preliminary action by pre-storing solutions to common SOP tasks in a database. When new tasks arise, the system retrieves relevant pre-computed solutions and adapts them rather than generating everything from scratch, reducing the computational burden and simplifying the real-time generation process.
Data Source
AI summary
A method and a system for using a large language model (LLM) to automatically translate textual instructions into executable software code via skill distillation and composition are provided. The method includes: receiving a request for performing a task and a prompt; providing, as an input to an LLM, the first request and a response to the prompt; receiving, from the LLM, a set of code that implements a function that corresponds to a skill that is usable for performing the task; generating a test that relates to the task; performing the test by executing the set of code and checking whether the task has been successfully completed; and when the task has been successfully completed, storing the set of code in a skills library. Sets of code stored in the skills library may then be accessed and combined in order to perform larger tasks.