MEMRES: A Memory-Augmented Resolver with Confidence Cascade for Agentic Python Dependency Resolution

2026-04-18

AI Summary

This work addresses the challenge of code execution failures in Python due to import-related dependency resolution errors by proposing a multi-level confidence cascaded intelligent agent system. The system prioritizes efficient rule-based reasoning and a self-evolving memory mechanism, invoking a large language model (Gemma-2 9B) only when necessary, thereby significantly enhancing both efficiency and accuracy. Key innovations include a semantic import analyzer, a Python 2 heuristic detector, and an error-pattern knowledge base integrating over 200 curated import-to-package mapping rules. Evaluated on the HG2.9K dataset, the method successfully resolves 2,503 out of 2,890 (86.6%) code snippets, substantially outperforming the 54.7% success rate of pure LLM-based approaches.

Technology Category

• Reasoning under Uncertainty: Probabilistic Programming
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Multiagent Systems: Multiagent Systems under Uncertainty

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present MEMRES, an agentic system for Python dependency resolution that introduces a multi-level confidence cascade where the LLM serves as the last resort. Our system combines: (1) a Self-Evolving Memory that accumulates reusable resolution patterns via tips and shortcuts; (2) an Error Pattern Knowledge Base with 200+ curated import-to-package mappings; (3) a Semantic Import Analyzer; and (4) a Python 2 heuristic detector resolving the largest failure category. On HG2.9K using Gemma-2 9B (10 GB VRAM). MEMRES resolves 2503 of 2890 (86.6%, 10-run average) snippets, combining intra-session memory with our confidence cascade for the remainder. This already exceeds PLLM's 54.7% overall success rate by a wide margin.

Problem

• dependency resolution
• Python imports
• error patterns
• package mapping
• code execution

Innovation

• Memory-Augmented Resolver
• Confidence Cascade
• Self-Evolving Memory
• Error Pattern Knowledge Base
• Semantic Import Analyzer