NEIL: Design of an AI-Driven Multi-Agent System for Accounts Payable Automation

Abstract

Accounts payable (AP) departments continue to rely on template-based optical character recognition tools, manual data entry with five-to-ten-per-cent error rates, and rule-based engines that scale poorly across vendors, languages, and tax jurisdictions. This paper presents the design of NEIL (Next-Generation Enterprise Invoice Learning), an AI-driven multi-agent platform built on a microservices architecture for end-to-end AP automation. The system separates concerns into two cooperating services: an Agent Service (Eddie) built on FastAPI and LangGraph that hosts specialised state-machine agents for document preprocessing, template-agnostic extraction using multimodal Large Language Models, validation with semantic matching, and tax-code mapping; and a Workflow Service (Neil) built on Django REST Framework, PostgreSQL, and Celery that handles multi-channel ingestion, invoice lifecycle, configurable approval routing, audit trails, and ERP integration. We describe the service decomposition, the inter-service REST contract, the Django app structure, the Celery task-queue topology, and the Kubernetes deployment posture that together enable straight-through processing. The architecture is positioned against the limitations of monolithic legacy AP platforms and is shown to support independent scalability, swappable AI providers, configuration-driven workflows, and observable agent state.

Introduction

The text discusses the importance of Human Resource Management (HRM) in improving employee engagement, especially in the context of globalization and digital transformation.

Employee engagement refers to the emotional, cognitive, and physical involvement of employees in their work, which leads to higher productivity, motivation, loyalty, and better organizational performance. The study emphasizes that HRM plays a central role in building engagement through key practices such as recruitment, training and development, performance management, compensation, communication, and work-life balance support.

It also highlights how digital transformation has both positive effects (better communication, flexibility, productivity) and challenges (burnout, isolation, stress), requiring HR strategies like virtual collaboration tools, wellness programs, and regular feedback systems.

The study identifies major challenges for HRM, including workforce diversity, retention issues, remote team management, and mental health concerns. To address these, organizations should focus on strong leadership, employee recognition, career development, inclusive culture, and effective use of technology.

Conclusion

This paper has described the design of NEIL, a microservices-based platform for AI-driven accounts payable automation. The key architectural choice is the deliberate separation between an Agent Service that hosts swappable LangGraph-based AI components and a Workflow Service that owns enterprise-facing lifecycle, audit, and ERP-integration responsibilities. The two services communicate through a small typed REST contract, are scaled independently on Kubernetes, and are designed so that either side can be evolved without disrupting the other. Future work proceeds in three directions. First, the integration of parameter-efficient fine-tuning workflows (LoRA [9], QLoRA [10]) into the Eddie deployment lifecycle for per-tenant adaptation of the extraction model. Second, the addition of an ensemble document-classifier (ResNet-50 [11], MobileNetV2 [12]) at the ingestion layer to triage incoming documents and route supporting documents away from the full extraction pipeline. Third, the construction of an externally-curated invoice-extraction benchmark drawn from anonymised production data, allowing direct empirical comparison between the proposed architecture and academic baselines under per-vendor and per-language reporting.

References

[1] G. Kim et al., “OCR-free Document Understanding Transformer (Donut),” in Proc. ECCV, 2022, pp. 498–517. [2] Y. Huang, T. Lv, L. Cui, Y. Lu, and F. Wei, “LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking,” in Proc. ACM MM, 2022, pp. 4083–4091. [3] M. Li et al., “TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models,” in Proc. AAAI, 2023, pp. 13094–13102. [4] Qwen Team, “Qwen2-VL: Enhancing Vision-Language Model’s Perception of the World at Any Resolution,” arXiv:2409.12191, 2024. [5] Meta AI, “The Llama 3 Herd of Models,” arXiv:2407.21783, 2024. [6] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks,” in Proc. EMNLP, 2019, pp. 3982–3992. [7] J. Johnson, M. Douze, and H. Jégou, “Billion-scale Similarity Search with GPUs,” IEEE Trans. Big Data, vol. 7, no. 3, pp. 535–547, 2021. [8] LangChain, “LangGraph: Building Stateful, Multi-Actor Applications with LLMs,” 2024. [Online]. Available: https://langchain-ai.github.io/langgraph/ [9] E. J. Hu et al., “LoRA: Low-Rank Adaptation of Large Language Models,” in Proc. ICLR, 2022. [10] T. Dettmers, A. Pagnoni, A. Holtzman, and L. Zettlemoyer, “QLoRA: Efficient Finetuning of Quantized LLMs,” in Proc. NeurIPS, 2023. [11] K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in Proc. IEEE CVPR, 2016, pp. 770–778. [12] M. Sandler et al., “MobileNetV2: Inverted Residuals and Linear Bottlenecks,” in Proc. IEEE CVPR, 2018, pp. 4510–4520. [13] Institute of Finance & Management (IOFM), “AP Department Benchmarks and Analysis,” annual industry survey, 2023.

Copyright

Copyright © 2026 Amaan A. Khan, D. P. Gaikwad. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.