MASAI: Modular Architecture for Software-engineering AI Agents

Nalin Wadhwa, Atharv Sonwane, Daman Arora, Abhav Mehrotra, Saiteja Utpala, Ramakrishna B Bairi, Aditya Kanade, Nagarajan Natarajan

Keywords: code generation and understanding, AI Agents

TL;DR: We propose a modular agent architecture that achieves competitive performance on software engineering tasks.

Abstract: A common method to solve complex problems in software engineering is to divide the problem into multiple sub-problems. Inspired by this, we propose a Modular Architecture for Software-engineering AI (MASAI) agents, where different LLM-powered sub-agents are instantiated with well-defined objectives and strategies tuned to achieve those objectives. Our modular architecture offers several advantages: (1) employing and tuning different problem-solving strategies across sub-agents, (2) enabling sub-agents to gather information from different sources scattered throughout a repository, and (3) avoiding unnecessarily long trajectories which inflate costs and add extraneous context. MASAI achieves a competitive performance (28.33% resolution rate) on the popular and highly challenging SWE-bench Lite dataset consisting of 300 GitHub issues from 11 Python repositories. at less than 2$ per issue cost on average.

Submission Number: 103