Digital Assistant Function Calling With Context-Aware LLM Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional digital assistants struggle with handling user inputs that deviate from predefined training data, lack flexibility in conversation flows, and have limited ability to integrate contextual data, leading to suboptimal performance and user experience.
Innovation Solution
A context-aware digital assistant leveraging a generative artificial intelligence (AI) model, such as a large language model (LLM), dynamically interprets user inputs to generate function calls or responses without explicit intent and entity definitions, enabling natural, context-aware conversations and modular function invocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If explicit conversation design with predefined intents and entities is used, then the digital assistant can handle predetermined conversation flows, but it cannot handle user inputs that deviate from training data
Solution Approach 1:
The system uses a large language model to automatically generate function calls from user inputs without requiring explicit intent classification or entity extraction. The LLM directly processes natural language inputs and generates appropriate function calls, eliminating the need for predefined conversation designs while maintaining reliable function execution.
Solution Approach 2:
The patent changes the fundamental parameter of conversation handling from structured intent-based processing to unstructured natural language processing. By using a large language model, the system transforms how user inputs are interpreted, allowing it to handle diverse inputs while maintaining conversation coherence through the model's contextual understanding capabilities.
2Stability of the object's composition
If predefined conversation flows are used, then the digital assistant can maintain structured interactions, but it cannot integrate contextual data items across messages dynamically
Solution Approach 1:
The system transitions from static predefined conversation flows to dynamic context-aware processing. The large language model maintains conversation context dynamically across multiple messages, adapting to user inputs in real-time while preserving coherent interactions. This allows the system to integrate contextual data items naturally without rigid structural constraints.
3Productivity
If explicit intent and entity modeling is implemented, then the digital assistant can execute specific actions, but it requires extensive training data and development effort
Solution Approach 1:
The patent extracts the complex intent classification and entity extraction components from the conversation processing pipeline. Instead of using separate modules for intent recognition and entity extraction, the system uses a large language model to directly generate function calls from natural language inputs, simplifying the architecture while maintaining action execution capability.
Data Source
AI summary
Example systems and methods described herein relate to a digital assistant service with automated function calling. Prompt data is provided to a generative machine learning model. The prompt data includes user input and function data. The user input is received via a user interface associated with a digital assistant. The function data identifies a plurality of functions. A response from the generative machine learning model includes a function identifier associated with a function from among the plurality of functions. In response to detecting that the response includes the function identifier, the function is invoked to obtain output data. The output data is caused to be presented in the user interface associated with the digital assistant.