LLM4FaaS: No-Code Application Development using LLMs and FaaS (2502.14450v1)

Abstract: LLMs are powerful tools that can generate code from natural language descriptions. While this theoretically enables non-technical users to develop their own applications, they typically lack the expertise to execute, deploy, and operate generated code. This poses a barrier for such users to leverage the power of LLMs for application development. In this paper, we propose leveraging the high levels of abstraction of the Function-as-a-Service (FaaS) paradigm to handle code execution and operation for non-technical users. FaaS offers function deployment without handling the underlying infrastructure, enabling users to execute LLM-generated code without concern for its operation and without requiring any technical expertise. We propose LLM4FaaS, a novel no-code application development approach that combines LLMs and FaaS platforms to enable non-technical users to build and run their own applications using only natural language descriptions. Specifically, LLM4FaaS takes user prompts, uses LLMs to generate function code based on those prompts, and deploys these functions through a FaaS platform that handles the application's operation. LLM4FaaS also leverages the FaaS infrastructure abstractions to reduce the task complexity for the LLM, improving result accuracy. We evaluate LLM4FaaS with a proof-of-concept implementation based on GPT-4o and an open-source FaaS platform, using real prompts from non-technical users. Our evaluation based on these real user prompts demonstrates the feasibility of our approach and shows that LLM4FaaS can reliably build and deploy code in 71.47% of cases, up from 43.48% in a baseline without FaaS.