Design and Implementation of Agentic AI Pipelines for Enterprise Decision-Making Architecture Patterns for Production Systems

Authors: Sandeep Nutakki

DOI: https://doi.org/10.37082/IJIRMPS.v14.i2.233041

Short DOI: https://doi.org/

Country: United States

Full-text Research PDF File:   View   |   Download

Abstract: The emergence of large language models (LLMs) has enabled a new paradigm of autonomous AI agents capable of reasoning, planning, and executing complex multi-step tasks. However, deploying these agents in enterprise environments presents significant architectural challenges around orchestration, tool integration, memory management, and execution reliability. This paper presents Nexus, a reference architecture and implementation for production-grade agentic AI pipelines in enterprise decision-making systems. We formalize reliability semantics (at-most-once, at-least-once, exactly-once) for LLM tool execution, introduce an adaptive reasoning strategy selector that dynamically chooses between Direct, ReAct, and Plan-and-Execute based on task complexity, and present a hierarchical memory system with explicit operational semantics. Evaluation across 500 real enterprise tasks demonstrates 87.6% end-to-end task completion rates with median latency of 1.2 seconds, achieving statistically significant improvements over fixed-strategy baselines (p < 0.01). Ablation studies validate the contribution of individual components. We analyze failure modes, discuss lessons learned from production deployments, and provide guidelines for practitioners building reliable agent systems.

Keywords: Agentic AI, Large Language Models, Pipeline Architecture, Enterprise AI, Autonomous Agents, Tool Integration, Production Systems.

Paper Id: 233041

Published On: 2026-04-27

Published In: Volume 14, Issue 2, March-April 2026

Share this