LLM4FaaS: No-Code Application Development using LLMs and FaaS

TL;DR

LLM4FaaS tackles the deployment bottleneck of LLM-generated code by integrating a no-code workflow with Function-as-a-Service, enabling non-technical users to build customized apps from natural language descriptions. The architecture combines an LLM for code generation, a FaaS deployment layer, and a bridging orchestrator that constructs prompts, parses outputs, and deploys functions end-to-end. Evaluations on real user descriptions show a semantic pass rate of 71.47% and a syntactic pass rate of 87.55%, with end-to-end latency around 23 seconds, illustrating feasible automation and practical performance. The work contributes a proof-of-concept prototype, public datasets, and cross-language analyses, highlighting a pathway to democratize application customization while maintaining code quality and availability.

Abstract

Large language models (LLMs) show great capabilities in generating code from natural language descriptions, bringing programming power closer to non-technical users. However, their lack of expertise in operating the generated code remains a key barrier to realizing customized applications. Function-as-a-Service (FaaS) platforms offer a high level of abstraction for code execution and deployment, allowing users to run LLM-generated code without requiring technical expertise or incurring operational overhead. In this paper, we present LLM4FaaS, a no-code application development approach that integrates LLMs and FaaS platforms to enable non-technical users to build and run customized applications using only natural language. By deploying LLM-generated code through FaaS, LLM4FaaS abstracts away infrastructure management and boilerplate code generation. We implement a proof-of-concept prototype based on an open-source FaaS platform, and evaluate it using real prompts from non-technical users. Experiments with GPT-4o show that LLM4FaaS can automatically build and deploy code in 71.47% of cases, outperforming a non-FaaS baseline at 43.48% and an existing LLM-based platform at 14.55%, narrowing the gap to human performance at 88.99%. Further analysis of code quality, programming language diversity, latency, and consistency demonstrates a balanced performance in terms of efficiency, maintainability and availability.

Figures (5)

• Figure 1: LLM4FaaS consists of three components: an LLM system, a FaaS system, and LLM-FaaS Bridge, the orchestration system. Together, they integrate the power of LLMs and the abstraction of FaaS to enable end-to-end application customization for non-technical users.
• Figure 2: Syntactic and semantic pass rates across four task complexities. We compare LLM4FaaS with (i) a baseline without FaaS integration, (ii) Open Interpreter, and (iii) human developer. GPT-4o is used for all LLM-based experiments.
• Figure 3: Comparison of LLM4FaaS syntactic and semantic pass rates between NodeJS and Python. Syntactic pass rates remain consistently high in both case, while the semantic pass rate decreases for all task complexities under the NodeJS setup.
• Figure 4: Average end-to-end latency of all experiments, decomposed into LLM generation, FaaS preparation, and deployment. Error bars indicate standard deviation.
• Figure 5: Pass rates of ten repetitions of LLM4FaaS with three user answers. Syntactic pass remains stable between repeat LLM invocations (nine or ten of ten correct). For the complex task, repeating an identical invocation can lead to different results.