FirmOS: Smart Intelligent Assistant for Chartered Accountant

Abstract

Chartered Accountant (CA) firms deal with a huge amount of client data and supporting paperwork for tax filing, audits, compliance and similar services. Gathering, checking, sorting and storing all of this by hand takes a lot of time, and mistakes or missing details often creep into reports and everyday work. This review look at the research behind FirmOS, an smart intelligence tool built for CA firms. It covers AI in accounting and auditing, optical character recognition (OCR), automated invoice processing, document understanding, natural language processing (NLP), semantic retrieval, retrieval-augmented generation (RAG), tool-using AI systems and human–AI collaboration. The studies we reviewed show real progress. OCR turns scanned pages into digital text, document AI models draw on both the words and the visual layout, semantic models retrieve information by meaning rather than keywords, and RAG links language models to outside data. Accounting research add that AI makes work more efficient but professional judgment is still needed. Still, most of this work treats each technology on its own and rarely looks at what a CA firm actually needs day to day. FirmOS tries to fill that gap by bringing together client information management, document collection, OCR-based data extraction, template-based verification, document status tracking and role-based access limited to authorized staff. It is meant to help CA professionals with their daily work, not to take their place.

Introduction

The research proposes FirmOS, an AI-powered practice management and intelligent assistant system for Chartered Accountant (CA) firms. The system aims to streamline client management, document collection, data extraction, verification, deadline tracking, and access to authorized information through natural-language queries. It combines existing AI and document-processing technologies into a unified workflow while ensuring that final professional decisions remain with qualified CA personnel.

1. Introduction and Objectives

CA firms manage large volumes of financial and compliance-related documents, including invoices, bank statements, tax records, receipts, and audit supporting documents. Manually collecting, verifying, organizing, and tracking these records consumes considerable time and can lead to missing documents, duplicated work, and missed deadlines.

FirmOS is proposed to address these challenges by centralizing firm activities and automating repetitive administrative tasks.

Its main objectives are to:

• Develop a centralized platform for managing clients, documents, and office activities.

• Simplify document collection, storage, and access.

• Automate text extraction and basic document verification using OCR.

• Track tasks, deadlines, missing documents, and work progress.

• Provide secure, role-based access to confidential client information.

• Improve employee productivity while retaining professional oversight.

2. Literature Review

The research reviews seven major areas that provide the technological and professional foundation for FirmOS.

• AI in accounting and auditing: Supports automation, financial analysis, risk assessment, anomaly detection, and audit assistance.

• OCR and document processing: Converts scanned documents into machine-readable text and supports automated invoice and financial-data extraction.

• Document AI: Technologies such as LayoutLM and LayoutLMv3 use textual and visual layout information to understand complex documents.

• Natural Language Processing (NLP): BERT and Sentence-BERT support language understanding, semantic similarity, and natural-language search.

• Retrieval-Augmented Generation (RAG): Combines information retrieval with language models to answer questions using relevant organizational records.

• Intelligent tool use: Research on Toolformer and ReAct explores how language models can interact with external tools and software functions.

• Human–AI collaboration: Emphasizes professional oversight, accountability, privacy, explainability, and verification of AI-generated results.

These studies establish that individual technologies can support accounting workflows, but they do not, by themselves, provide a complete practice management solution tailored to CA firms.

3. Research Gap and Problem Statement

Existing research generally focuses on individual capabilities, such as OCR, invoice extraction, document understanding, or AI-assisted question answering. CA firms, however, require these capabilities to work together with client records, document requirements, task management, deadlines, and professional review.

The identified research gaps include:

• Lack of an integrated workflow designed specifically for CA practices.

• Separation between document extraction and client or practice management.

• Need for secure, permission-based retrieval of confidential information.

• Need for human verification and professional accountability.

• Lack of practical integrated prototypes demonstrating how these capabilities can work together.

FirmOS addresses this application-level gap by integrating established technologies into a centralized system for managing client-related work.

4. Proposed Methodology

The proposed system follows an integrated workflow:

1

Client Management

Maintain client profiles and define required documents.

2

Client Workspace

Organize client-specific records and document uploads.

3

Document Processing

Use OCR to extract information from scanned and image-based files.

4

Verification and Organization

Check extracted details against predefined rules and categorize documents.

5

Status and Deadline Tracking

Identify missing documents, pending tasks, and approaching deadlines.

6

Natural Language Assistant

Retrieve relevant authorized information in response to plain-language questions.

7

CA Review and Decision

Allow qualified professionals to verify results and make final decisions.

The workflow connects document processing with client-specific information and task tracking. Extracted information is checked against predefined rules, while missing or inconsistent details are flagged for review. A natural-language assistant can help authorized users find relevant records without manually searching through numerous files.

The final stage is professional review, where the CA or authorized employee validates important results before taking action.

5. Conceptual Framework

FirmOS brings together three broad layers:

• Document intelligence: OCR, document extraction, classification, and verification.

• Practice management and information retrieval: Client workspaces, task tracking, deadlines, permissions, semantic search, and potentially RAG-based question answering.

• Professional oversight: Human review, accountability, security, and final decision-making.

Together, these layers aim to transform disconnected administrative activities into a more organized and intelligent workflow.

6. Security and Human–AI Collaboration

Because CA firms handle confidential financial and personal information, the proposed system must incorporate authentication, role-based access, protected document storage, and authorized information retrieval.

AI should assist with repetitive tasks and information discovery rather than independently making professional accounting or auditing decisions. Extracted data, verification flags, and generated answers must be checked where necessary.

The research also recognizes that advanced semantic retrieval, RAG, layout-aware document understanding, and controlled AI tool use may require further development. The implemented prototype should be distinguished clearly from proposed future capabilities.

Conclusion

This literature review examined research relevant to an smart intelligence assistant for Chartered Accountant firms. The literature covered AI in accounting and auditing, OCR, automated invoice processing, document understanding, NLP, semantic representation, retrieval, RAG, intelligent tool use and human–AI collaboration. Together, these studies provide a technical and professional basis for FirmOS. The review also reveals a significant gap at the application level. Studies to focus on particular technologies or individual accounting tasks, where as a CA practice needs connected management of clients, documents, requirements, verification, status and information retrieval. What is needed is an integrated workflow that links document intelligence with practice management and controlled access to organization-specific information. FirmOS is put forward as an application-oriented answer to this gap. Its core workflow combines client management, document collection, OCR based extraction, predefined verification, organization, status tracking and natural-language access to authorized information. Advanced Document AI, semantic retrieval, RAG and controlled tool use can be seen as future extensions. Throughout the design, human in the loop review remains essential, so that AI handles repetitive and information processing work while professional CA judgment stays the final decision layer.

References

[1] A. M. Abdo-Salloum and S. Chehade, “The Role of Artificial Intelligence in Transforming Accounting and Auditing Practices: A Systematic Review,” SAGE Open, 2026, doi: 10.1177/21582440251403296. [2] R. Smith, “An Overview of the Tesseract OCR Engine,” in Proc. Ninth International Conference on Document Analysis and Recognition (ICDAR), 2007, pp. 629–633, doi: 10.1109/ICDAR.2007.4376991. [3] Y. Xu, M. Li, L. Cui, S. Huang, F. Wei, and M. Zhou, “LayoutLM: Pre-training of Text and Layout for Document Image Understanding,” in Proc. 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2020, doi: 10.1145/3394486.3403172. [4] 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. 30th ACM International Conference on Multimedia, 2022, pp. 4083–4091, doi: 10.1145/3503161.3548112. [5] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks,” in Proc. EMNLP-IJCNLP, 2019, pp. 3982–3992, doi: 10.18653/v1/D19-1410. [6] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in Proc. NAACL-HLT, 2019, pp. 4171–4186, doi: 10.18653/v1/N19-1423. [7] M. M. Dong, T. C. Stratopoulos, and V. X. Wang, “A Scoping Review of ChatGPT Research in Accounting and Finance,” International Journal of Accounting Information Systems, vol. 55, 100715, 2024, doi: 10.1016/j.accinf.2024.100715. [8] F. Krieger, P. Drews, and B. Funk, “Automated invoice processing: Machine learning-based information extraction for long tail suppliers,” Intelligent Systems with Applications, vol. 20, 200285, 2023, doi: 10.1016/j.iswa.2023.200285. [9] M. Mezzara, A. Saoudi, and J. Zahi, “Process Efficiency in Moroccan Accounting Firms: A Qualitative Study and an OCR-Based Automation Proposal,” Procedia Computer Science, vol. 280, pp. 1278–1283, 2026, doi: 10.1016/j.procs.2026.04.171. [10] C. Wang, J. Zhang, and Y. Yu, “Invoiveocr: Deep Learning Based Automatic Invoice Recognition Process,” Discover Artificial Intelligence, vol. 6, article 636, 2026, doi: 10.1007/s44163-026-01140-3. [11] T. Schick et al., “Toolformer: Language Models Can Teach Themselves to Use Tools,” 2023, arXiv:2302.04761, doi: 10.48550/arXiv.2302.04761. [12] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” in Advances in Neural Information Processing Systems, 2020, arXiv:2005.11401, doi: 10.48550/arXiv.2005.11401. [13] Y. Gao et al., “Retrieval-Augmented Generation for Large Language Models: A Survey,” 2024, arXiv:2312.10997, doi: 10.48550/arXiv.2312.10997. [14] J. H. Choi and C. L. Xie, “Human + AI in Accounting: Early Evidence from the Field,” Journal of Accounting Research, vol. 64, no. 3, pp. 1333–1373, 2026, doi: 10.1111/1475-679X.70052. [15] P. Rikhardsson, C. Batt, K. R. Thórisson, and G. Bergthorsson, “Artificial intelligence and auditing in small- and medium-sized firms: Expectations and applications,” AI Magazine, vol. 43, pp. 323–336, 2022, doi: 10.1002/aaai.12066. [16] T. Sun and M. A. Vasarhelyi, “Embracing Textual Data Analytics in Auditing with Deep Learning,” International Journal of Digital Accounting Research, vol. 18, pp. 49–67, 2018, doi: 10.4192/1577-8517-v18_3. [17] K. Guu, K. Lee, Z. Tung, P. Pasupat, and M.-W. Chang, “REALM: Retrieval-Augmented Language Model Pre-Training,” in Proc. ICML, 2020, PMLR 119, arXiv:2002.08909, doi: 10.48550/arXiv.2002.08909. [18] V. Karpukhin et al., “Dense Passage Retrieval for Open-Domain Question Answering,” in Proc. EMNLP, 2020, pp. 6769–6781, doi: 10.18653/v1/2020.emnlp-main.550. [19] S. Yao et al., “ReAct: Synergizing Reasoning and Acting in Language Models,” in Proc. ICLR, 2023, arXiv:2210.03629, doi: 10.48550/arXiv.2210.03629. [20] A. Zemankova, “Artificial Intelligence in Audit and Accounting: Development, Current Trends, Opportunities and Threats – Literature Review,” in Proc. 2019 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO), 2019, pp. 148–154, doi: 10.1109/ICCAIRO47923.2019.00031.

Copyright

Copyright © 2026 Prof. Akshata H. Wagh, Mr. Aayush P. Lodhe, Mr. Kaustubh K. Deshmukh, Mr. Manoj S. Ahire, Mr. Suraj B. Hinde. 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.