A method and system for automated processing of environmental, social and governance disclosure documents

Abstract

The present invention discloses a computer-implemented system and method for automated evaluation of Environmental, Social, and Governance (ESG) disclosures and detection of greenwashing. The system integrates artificial intelligence and machine learning techniques to extract, segment, and analyze ESG content from structured and unstructured sources. A retrieval and segmentation module identifies ESG-relevant sections, which are then vectorized and stored for semantic search. A semantic retrieval module fetches contextually relevant content, which is evaluated against predefined benchmarks using an analysis and validation module. The scoring module assigns ESG scores based on rule-based criteria including key strengths, specificity, and supporting evidence. A report generation module creates structured ESG reports with visual analytics, while an orchestrator module dynamically coordinates all processing steps based on ESG framework requirements.

Description

A METHOD AND SYSTEM FOR AUTOMATED PROCESSING OF ENVIRONMENTAL, SOCIAL AND GOVERNANCE DISCLOSURE DOCUMENTSFIELD OF THE INVENTION

[0001] The present invention relates to Environmental, Social, and Governance (ESG) disclosure assessment, and particularly, to a method and system for ESG evaluation and greenwashing detection using disclosure -performance mapping, regulatory compliance analysis, and benchmarking-driven inconsistency identification.BACKGROUND OF THE INVENTION

[0002] Background description includes information that may be useful in understanding the present invention.

[0003] ESG metrics have become pivotal parameters for evaluating the long-term sustainability and ethical practices of global corporate operations. Increasingly, stakeholders such as investors, customers, regulatory bodies, and the general public rely on ESG disclosures to assess corporate responsibility and sustainability. However, the measurement and reporting of ESG data remain highly dynamic, creating strategic and operational uncertainty for corporate leaders as they strive to prepare disclosures aligned with evolving expectations and communicate them effectively to diverse audiences.

[0004] One core challenge lies in the absence of standardized global frameworks for evaluating ESG-related performance. The fragmented and inconsistent nature of ESG reporting standards has led to confusion and difficulty in comparing disclosures across companies, sectors, and jurisdictions. Enterprises operating across international borders often question the feasibility of universal sustainability metrics, while attempting to balance divergent stakeholder priorities ranging from profit-focused shareholders to sustainability-conscious consumers and regulators. Additionally, organizations are often forced to manage stakeholder perception in polarized environments where ESG efforts may be both lauded and scrutinized.

[0005] To address such concerns, individual regions have developed localized ESG reporting standards. In India, the Securities and Exchange Board of India (SEBI) has mandated the Business Responsibility and Sustainability Reporting (BRSR) framework for the top 1,000 listed companies beginning in fiscal year 2022-23. This initiative represents India's distinctive approach to ESG reporting and is built around nine core principles, including business ethics, environmental responsibility, human rights, employee well-being, and consumer value. Despite its potential to promote transparency and comparability, BRSR adoption has not entirely resolved challengesrelated to reliability, standardization, and accountability in sustainability disclosures.

[0006] Compounding these issues, companies worldwide face inconsistencies in how ESG ratings are assigned. ESG scores may be based on either solicited assessments, where companies request evaluations, or unsolicited evaluations conducted independently by agencies such as MSCI ESG Research, Sustainalytics, S&P Global, Bloomberg, and Refinitiv. Each agency employs its own distinct methodology, resulting in divergent outcomes for the same company. This misalignment undermines the credibility of ESG scores and creates confusion for stakeholders attempting to assess a company's true sustainability performance.

[0007] In the Indian context, while BRSR reports undergo third-party audits and encompass Scope 1 (direct emissions), Scope 2 (indirect emissions from purchased energy), and Scope 3 (valuechain-related emissions), inconsistencies still persist. Firms may unintentionally or deliberately publish disclosures containing data inaccuracies, exaggerated claims, or misleading narratives-a phenomenon commonly referred to as greenwashing. Despite structured frameworks and audit mechanisms, companies may distort ESG-related information, especially where Scope 3 emissions are concerned. Estimating these emissions requires data across a company's entire supply chain, including supplier behavior, product usage, and end of-life outcomes data which is often hard to obtain and lacks standardization in calculation methodologies.

[0008] The push for ESG compliance also presents economic and organizational hurdles. Many firms struggle to afford sustainability investments, such as transitioning to renewable energy or adopting low-emission technologies. Stakeholder interests may also diverge: while external stakeholders demand responsible corporate conduct, internal shareholders may prioritize shortterm profitability, resulting in conflicting pressures. Regulatory mandates often incentivize sustainability reporting, yet companies may opt to demonstrate compliance without meaningful change further raising concerns about the credibility of ESG commitments.

[0009] With ESG disclosures influencing capital allocation, investor trust, and market valuation, the risks associated with misrepresentation have grown substantially. Companies are increasingly motivated to enhance their perceived ESG performance, which may result in overstated or selectively disclosed sustainability data. Greenwashing, in particular, has emerged as a significant threat to transparency and investor confidence. Common manifestations include the omission of negative environmental data, deceptive language in product claims, or inflated representations of ESG achievements in formal disclosures. This behavior often results in ESG disclosure ratings that far exceed a company's actual operational performance a discrepancy quantifiable through metrics such as greenwashing scores.

[0010] Additionally, firm-level behavioral phenomena such as information asymmetry, green hushing (the practice of deliberately withholding ESG efforts to avoid scrutiny), blue-washing (overemphasizing social impact), and carbon washing (manipulating carbon-related claims) have further complicated ESG evaluation. These tactics allow organizations to craft ESG narratives that may not reflect ground realities. Studies in India, for example, have observed higher instances of greenwashing among NIFTY 50 companies in the manufacturing and energy sectors compared to lower-risk sectors like IT and FMCG. These inconsistencies not only erode stakeholder trust but also hamper the establishment of reliable ESG benchmarks across industries.

[0011] Despite these multifaceted challenges, ESG evaluation remains a critical area of research, with direct implications on corporate accountability, regulatory compliance, and investor protection. Although frameworks like BRSR offer structure, they do not guarantee verifiable accuracy of ESG disclosures. As a result, there is growing urgency to develop advanced tools and systems that can rigorously assess ESG claims, detect inconsistencies, and provide transparency to stakeholders and regulators alike.

[0012] Therefore, there is a need for a method and system for ESG evaluation and greenwashing detection that enables standardized assessment of ESG disclosures, identifies discrepancies between reported and actual sustainability performance, and ensures the integrity of corporate sustainability claims through automated, Al-powered, and benchmark-driven validation mechanisms.

[0013] Some of the objects of the present disclosure, which at least one embodiment herein satisfies, are listed herein below.

[0014] The primary object of the present invention is to provide a system and method for ESG evaluation and greenwashing detection that integrates AI / ML-drivcn analytics with disclosure performance mapping and regulatory compliance verification to assess ESG disclosures against established standards, including BRSR.

[0015] Another object of the present invention is to provide an ESG evaluation module configured to extract and analyze ESG disclosures from structured and unstructured sources using Al and ML models, with particular emphasis on detecting misleading 'green language,' overgeneralized statements, and other textual indicators of greenwashing.

[0016] Another object of the present invention is to provide a greenwashing detection module employing unsupervised learning techniques to identify hidden patterns, anomalies, and outliersin ESG data that may signify deceptive practices, including selective disclosure and exaggeration of sustainability performance.

[0017] Another object of the present invention is to generate a projected ESG score by benchmarking disclosure content against industry standards, using feedback components such as: (a) identification of greenwashed content, (b) improvement recommendations, and (c) classification of disclosures as Laggard, Average, or Leader based on relative performance.

[0018] Another object of the present invention is to ensure that the evaluation framework incorporates all ESG parameters defined under India's mandatory BRSR framework, including sustainability practices mapped to the nine principles-ranging from ethics and employee welfare to environmental protection and consumer value.

[0019] Another object of the present invention is to provide a compliance verification module that assesses alignment of ESG disclosures with applicable regulatory and sectoral standards, such as those defined by SEBI, GRI, SASB, and the BRSR framework.

[0020] Another object of the present invention is to enable end-users to iteratively improve ESG disclosures by offering automated feedback based on Al-driven linguistic and semantic analysis, along with real-time visualizations of greenwashing probability and content-level scoring.

[0021] Another object of the present invention is to automate ESG report generation through intelligent parsing and data extraction from cloud-based storage systems, supporting scalability across sectors and standardizing disclosures according to global and regional norms.

[0022] Another object of the present invention is to enable investors, stakeholders, and regulatory bodies to make informed decisions by providing credible, verifiable, and benchmarked ESG assessments that mitigate information asymmetry and counter greenwashing risks.

[0023] Another object of the present invention is to provide a comparative benchmarking module configured to contextualize a firm's ESG disclosures relative to peers within the same industry, highlighting gaps and best practices based on aggregated industry data and ESG maturity levels.

[0024] Another object of the present invention is to create a multi-tiered scoring and classification structure that reflects sector- specific ESG compliance and sustainability maturity, enabling classification of firms into Laggard, Average, and Leader categories.

[0025] Another object of the present invention is to detect sector- specific greenwashing practices by leveraging Al models trained on historical data from industries with varying levels of greenwashing tendencies, such as manufacturing, energy, FMCG, and IT.

[0026] Another object of the present invention is to facilitate transparency and accountability inESG reporting by identifying and quantifying inconsistencies between ESG disclosure language and actual sustainability performance using explainable Al models.

[0027] Another object of the present invention is to support ESG disclosure refinement and integrity by generating final ESG evaluation reports that include both a projected ESG score and a quantified greenwashing score, empowering firms to align communication with actual practices.

[0028] Another object of the present invention is to offer a scalable and modular ESG evaluation and greenwashing detection system suitable for deployment by firms, investors, rating agencies, and regulators across geographies and sectors.

[0029] These and other objects and advantages will become more apparent when reference is made to the following description and accompanying drawings.SUMMARY OF THE INVENTION

[0030] This summary is provided to introduce concepts related to ESG disclosure assessment, and particularly, to a method and system for ESG evaluation and greenwashing detection using disclosure-performance mapping, regulatory compliance analysis, and benchmarking -driven inconsistency identification. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

[0031] In an aspect of the present invention, a computer-implemented method for automated processing of ESG disclosure documents are describes. The method includes the step of segmenting, by a processor using a retrieval and segmentation module, ESG-specific content sections from one or more ESG disclosure documents retrieved, by the processor using the retrieval and segmentation module, from at least one first source. The method further includes the step of converting, by the processor using a vectorization and storage module, the segmented content into vector representations and storing the vectors in a searchable vector database. The method further includes the step of performing, by the processor using a semantic retrieval module, semantic search over the vector database based on ESG-specific queries to retrieve contextually relevant disclosure content. The method further includes the step of evaluating, by the processor using an analysis and validation module, the retrieved content against predefined ESG standards and validating the authenticity based on external benchmark data obtained, by the processor using an analysis and validation module, from at least one second source. The method further includes the step of assigning, by the processor using a scoring module, ESG scores to the disclosure documents based on the evaluation and validation results based on a rule -based or weighted scoring mechanism. The method further includes the step of generating, by the processor using areport generation module, a structured, humanreadable ESG report comprising ESG performance metrics and visual outputs, based on the scoring and validation results. Finally, the method includes the step of dynamically coordinating, by the processor using an orchestrator module, the sequential steps segmenting to generating. The orchestrator module is configured to manage intermodule data flows based on content type of the disclosure documents or ESG reporting framework.

[0032] In an embodiment of the present invention, the scoring module applies a 100-point evaluation comprising four weighted criteria: key strengths, improvement suggestions, specificity, and supporting evidence.

[0033] In another embodiment of the present invention, the analysis and validation module dynamically formulates queries based on the sector and operational domain of the disclosing entity.

[0034] In another embodiment of the present invention, the analysis and validation module detects vague or unverifiable claims indicative of greenwashing, based on mismatches between internal disclosures and external data.

[0035] In another embodiment of the present invention, the retrieval and segmentation module preserves structural elements such as headings, tables, and metadata during extraction.

[0036] In another embodiment of the present invention, the segmentation is performed based on a predefined taxonomy of ESG principles comprising principles Pl to P9.

[0037] In another embodiment of the present invention, the vector representations are computed using deep learning-based models and stored in the searchable vector database implemented based on a similarity search engine comprising Facebook Al Similarity Search (FAISS).

[0038] In another embodiment of the present invention, the step of retrieving includes the sub-step of ranking results based on contextual relevance and identifying supporting evidence for each ESG content.

[0039] In another embodiment of the present invention, the evaluation is conducted based on industry-specific benchmarks. The industry-specific benchmarks includes the Global Reporting Initiative (GRI), Sustainability Accounting Standards Board (SASB), and the Task Force on Climate-related Financial Disclosures (TCFD).

[0040] In another embodiment of the present invention, the report generation module produces the visual outputs including radar charts, heat maps, and benchmark comparisons. The report generation module interfaces with a dashboard to enable user-driven interaction and feedback incorporation.

[0041] In another embodiment of the present invention, the step of dynamically coordinating, bythe processor using the orchestrator module, includes the sub-step of maintaining a framework identification knowledge base that is continually updated with pattern recognition models trained on structural and semantic characteristics of each ESG reporting framework. The step of dynamically coordinating further includes the sub-step of applying framework- specific error detection thresholds, wherein the Global Reporting Initiative (GRI) framework processing implements section-completeness validation, the Sustainability Accounting Standards Board (SASB) framework processing enforces industry-specific disclosure requirements, and the Task Force on Climate -related Financial Disclosures (TCFD) framework processing verifies the inclusion of forward-looking climate scenario analyses. The step of dynamically coordinating further includes the sub-step of tracking framework evolution by maintaining version -specific processing rules for each ESG standard and automatically identifying the applicable framework version based on document metadata. Finally, the step of dynamically coordinating includes the sub-step of implementing hybrid processing paths for multi-framework disclosures by detecting overlapping framework requirements and reconciling conflicting disclosure expectations through materiality-based prioritization techniques.

[0042] In another aspect of the present invention, a system for automated processing of ESG disclosure documents is described. The system includes a retrieval and segmentation module, a vectorization and storage module, a semantic retrieval module, an analysis and validation module, a scoring module, a report generation module, and an orchestrator module. The retrieval and segmentation module is configured to retrieve one or more ESG disclosure documents from at least one first source, and segment ESG-specific content sections from the retrieved disclosure documents. The vectorization and storage module is configured to generate vector representations of the segmented content and store them in a searchable vector database. The semantic retrieval module is configured to retrieve contextually relevant disclosure content based on ESG-specific queries from the vector database. The analysis and validation module is configured to assess retrieved content against predefined ESG standards and validate the authenticity based on external benchmark data obtained from at least one second source. The scoring module is configured to assign ESG scores based on the evaluation and validation results using a rule -based or weighted scoring mechanism. The report generation module is configured to produce structured, human- readable reports comprising ESG performance metrics and visual outputs. The orchestrator module is configured to coordinate the sequential operations of the preceding modules, and to handle intermodule data flows dynamically based on content type of the disclosure documents or ESG reporting framework.

[0043] Various objects, features, aspects, and advantages of the inventive subject matter willbecome more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which like numerals represent like components.BRIEF DESCRIPTION OF FRAWINGS

[0044] The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example and simply illustrates certain selected embodiments of devices, apparatus, systems, and methods that are consistent with the subject matter as claimed herein, wherein:

[0045] FIG. 1 illustrates a schematic block diagram of a Document Processing Agent for transforming ESG disclosure documents into machine-readable formats through OCR processing, text extraction, and structural preservation, in accordance with an exemplary embodiment of the present disclosure;

[0046] FIG. 2 illustrates a functional diagram of a Vision-based OCR Agent for converting scanned documents and PDFs into structured text while preserving formatting, tables, and visual elements, in accordance with an exemplary embodiment of the present disclosure;

[0047] FIG. 3 illustrates a schematic view of the Segmentation Agent implemented by the retrieval and segmentation module for organizing extracted content based on ESG principles Pl to P9, in accordance with an exemplary embodiment of the present disclosure;

[0048] FIG. 4 illustrates a process flow diagram of the Vectorization Agent implemented by the vectorization and storage module to convert segmented content into vector representations and store them in a searchable vector database, in accordance with an exemplary embodiment of the present disclosure;

[0049] FIG. 5 illustrates a block diagram of the Web Researching Agent supporting the analysis and validation module in retrieving benchmark data from at least one second source, in accordance with an exemplary embodiment of the present disclosure;

[0050] FIG. 6 illustrates a functional diagram of the Retrieval Agent using the semantic retrieval module to perform semantic search and retrieve contextually relevant ESG disclosure content, in accordance with an exemplary embodiment of the present disclosure;

[0051] FIG. 7 illustrates a system component view of the ESG Principle Analysis Agent using the analysis and validation module to evaluate disclosure content against predefined ESG standards, in accordance with an exemplary embodiment of the present disclosure;

[0052] FIG. 8 illustrates a block diagram of the ESG Scoring Agent using the scoring module to assign ESG scores using a 100-point rule -based or weighted evaluation mechanism, in accordance with an exemplary embodiment of the present disclosure;

[0053] FIG. 9 illustrates a schematic view of the Overall Feedback Agent implemented by the orchestrator module to dynamically coordinate inter-module operations and apply framework specific processing, in accordance with an exemplary embodiment of the present disclosure;

[0054] FIG. 10 illustrates a functional flow diagram of the Report Generation Agent using the report generation module to generate structured, human-readable ESG reports with performance metrics and visual outputs, in accordance with an exemplary embodiment of the present disclosure;

[0055] FIG. 11 illustrates a system workflow diagram showing coordinated operation of all modules from document ingestion to ESG insight generation using a multi-agent framework, in accordance with an exemplary embodiment of the present disclosure;

[0056] FIG. 12 illustrates a schematic overview of the Multi-Agent Framework showing six specialized agents arranged in a hub-and-spoke architecture centered around the orchestrator module, in accordance with an exemplary embodiment of the present disclosure;

[0057] FIG. 13 illustrates a block diagram of the system hardware for automated processing of ESG disclosure documents comprising processor, memory, communication interface, retrieval and segmentation module, vectorization and storage module, semantic retrieval module, analysis and validation module, scoring module, and orchestrator module, in accordance with an exemplary embodiment of the present disclosure; and

[0058] FIG. 14 illustrates a method flow diagram for the computer-implemented method of automated processing of ESG disclosure documents executed by the processor using respective modules, in accordance with an exemplary embodiment of the present disclosure.

[0059] The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.DESCRIPTION OF THE INVENTION

[0060] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to clearly communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, andalternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.

[0061] While the embodiments of the disclosure are subject to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the figures and will be described below. It should be understood, however, that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.

[0062] The terms “comprises”, “comprising”, or any other variations thereof used in the disclosure, are intended to cover a non-exclusive inclusion, such that a device, system, or assembly that comprises a list of components does not include only those components but may include other components not expressly listed or inherent to such system, or assembly, or device. In other words, one or more elements in a system or device proceeded by “comprises... a” does not, without more constraints, preclude the existence of other elements or additional elements in the system or device.

[0063] The present invention relates to a computer- implemented system and method for automated processing and evaluation of ESG disclosure documents, particularly in the context of sustainability reporting frameworks such as the BRSR requirements prescribed by the SEBI. The system is designed to enhance the transparency, accuracy, and integrity of ESG disclosures through the application of advanced artificial intelligence (Al), machine learning (ML), and multiagent system (MAS) architectures.

[0064] In a preferred embodiment, the system includes a modular, orchestrated pipeline of interoperable agents or modules that perform segmentation, vectorization, semantic retrieval, evaluation, scoring, and reporting of ESG content. Each module in the system is executed by a processor configured to perform specific sub-tasks, which are dynamically coordinated by an orchestrator module. The architecture supports adaptability to different ESG reporting frameworks, including but not limited to the GRI (Global Reporting Initiative), SASB (Sustainability Accounting Standards Board), and TCFD (Task Force on Climate-related Financial Disclosures), in addition to the Indian BRSR guidelines.

[0065] In accordance with an embodiment of the invention, a retrieval and segmentation module is configured to retrieve ESG disclosure documents from at least one first source, including cloud storage repositories, regulatory filing databases (e.g., PROWESS, SEBI, NSE, BSE), or third-party ESG platforms. The module further segments ESG-specific content based on structural elements (e.g., section headers, tables, figures, and metadata), preserving document hierarchy and semantic coherence. Segmentation is performed according to a predefined taxonomy of ESG principles,such as those defined in the National Guidelines on Responsible Business Conduct (NGRBC), encompassing nine core principles (Pl to P9) of corporate sustainability.

[0066] The segmented ESG content is transformed, by a vectorization and storage module, into high-dimensional vector embeddings using deep learning -based language models fine-tuned on ESG domain corpora. These vector representations are stored in a searchable vector database implemented on similarity search frameworks such as FAISS (Facebook Al Similarity Search), which supports rapid and scalable similarity matching across disclosure content.

[0067] To support vectorization and semantic retrieval of ESG-specific content, the system utilizes transformer-based encoder models leveraging bidirectional attention mechanisms, primarily based on BERT-style architectures. Sentence embedding models optimized for semantic similarity tasks, such as Sentence-BERT variants, are employed to generate contextual vector representations for ESG disclosure documents. These embeddings are refined using multi-layer perceptron projection heads, reducing dimensionality from 768 to 384 or 512 as required. For processing long-form ESG documents that exceed standard token limits, hierarchical attention networks are utilized.

[0068] The models undergo a two-stage fine-tuning process. In the continued pre-training phase, transformer models are exposed to large ESG-specific corpora using unsupervised objectives, thereby aligning internal representations with domain-specific semantics. This is followed by supervised fine-tuning using contrastive learning on curated ESG disclosure similarity pairs. Multi-task optimization is applied to jointly train classification heads for ESG category prediction alongside embedding generation. To ensure parameter-efficient training, techniques such as adapter layers and Low-Rank Adaptation (LoRA) are adopted, minimizing computational overhead while retaining performance.

[0069] Domain adaptation includes tokenizer expansion to incorporate ESG-relevant vocabulary, acronyms, and structured domain n-grams. Specialized attention mechanisms are configured to assign learned weights to ESG-critical keywords and numerical tokens, such as emissions data, financial figures, and percentage -based sustainability metrics. Furthermore, the architecture includes multi-lingual support to process ESG disclosures across diverse regulatory jurisdictions.

[0070] Training is conducted on a large-scale corpus including 50 to 200 million ESG-related text segments sourced from corporate sustainability reports, CDP Climate Change responses, GRL compliant documents, SASB filings, and regulatory submissions. The dataset spans 10 to 15 years of ESG disclosure history across multiple geographies, including SEC (US), NFRD (EU), and emerging market filings. Documents adhere to professional editorial standards and are tagged with structured metadata such as industry classifications (e.g., GICS / NAICS), jurisdictional tags, andframework identifiers. The data pipeline incorporates quality filtering to remove duplicates and low-information content while maintaining stratified representation across ESG themes and sectors.

[0071] For similarity search over the vector database, the system employs FAISS with IVFPQ (Inverted File with Product Quantization) indexing. This configuration supports over 10A8 vector operations and delivers sub- 100ms semantic query responses. K-means clustering is used for locality optimization to improve retrieval accuracy and system scalability.

[0072] A semantic retrieval module enables ESG-specific semantic search across the vector database using natural language queries derived from user prompts, regulatory requirements, or predefined templates. This module ranks the retrieved results based on contextual relevance and highlights supporting evidence from the source documents. This allows users and evaluators to identify both direct responses to ESG requirements and relevant auxiliary content that may substantiate or contradict reported claims.

[0073] In another embodiment, the system includes an analysis and validation module that evaluates retrieved disclosure content against predefined ESG standards and authenticates claims using external benchmark data retrieved from at least one second source. Such second sources include third-party ESG rating providers, government databases, environmental audits, carbon disclosure platforms, and publicly available operational data. The module dynamically formulates queries based on the sectoral classification and operational domain of the disclosing entity, thereby enabling industry-specific benchmarking and evaluation.

[0074] A key functionality of the analysis and validation module is greenwashing detection. The module detects vague, unverifiable, or contradictory claims by identifying semantic discrepancies between the disclosed ESG narrative and the independently verified facts. A hybrid detection strategy is implemented using supervised ML models trained on labeled BRSR reports (including genuine and greenwashed content) and unsupervised anomaly detection techniques for identifying latent patterns of exaggeration, omission, or misrepresentation. Claims / declarations / statements flagged as potentially greenwashed are further assessed using confidence scoring metrics and contextual analysis based on Indian sectoral norms.

[0075] Upon validation, a scoring module computes an ESG performance score for the disclosure documents based on a rule-based or weighted scoring mechanism. In one embodiment, the module employs a 100-point evaluation schema comprising four weighted criteria: (a) key strengths, (b) improvement suggestions, (c) specificity of disclosures, and (d) supporting evidence quality. This score reflects both the accuracy of ESG content and its alignment with applicable frameworks andbenchmarks.

[0076] Based on the scores and validation results, a report generation module synthesizes a structured, human-readable ESG report. The report includes performance metrics, visual analytics such as radar charts, heat maps, and comparative benchmarks, and a comprehensive evaluation of the level of greenwashing. The module interfaces with a web-based or platformnative dashboard that supports user interaction, feedback incorporation, and iterative disclosure refinement. The users can revise flagged content and regenerate updated reports reflecting improved transparency.

[0077] The entire sequence of operations is managed by an orchestrator module that dynamically coordinates inter-module data flows and ensures compliance with specific ESG reporting frameworks. The orchestrator maintains a knowledge base of ESG frameworks and utilizes AI- based pattern recognition models to classify the disclosure document structure and content. It applies framework- specific error detection thresholds, such as completeness checks, sectorspecific enforcement, and scenario analysis validation, specifically the Global Reporting Initiative (GRI), Sustainability Accounting Standards Board (SASB), and Task Force on Climate-related Financial Disclosures (TCFD), respectively, and identifies the applicable framework version using metadata parsing.

[0078] Furthermore, the orchestrator module supports hybrid processing of multi -framework disclosures by identifying overlapping ESG requirements and resolving conflicts through materiality-based prioritization algorithms. This functionality enables the system to generate a unified, reconciled evaluation output even for complex multi-standard documents, particularly relevant for multinational corporations reporting across jurisdictions.

[0079] In a specialized embodiment for Indian regulatory compliance, the system is trained on annotated datasets of BRSR reports from NIFTY 100 companies, with reference to India specific standards such as BIS norms, NGRBC principles, and sectoral sustainability guidelines. The use of such tailored datasets enhances the model's domain understanding and ensures precise assessment of BRSR adherence, making the invention particularly impactful in the Indian ESG landscape.

[0080] The system further includes modules for advanced feature engineering and model training. Feature engineering captures linguistic markers of greenwashing, including vague language, hyperbolic assertions, and selective omission. Semantic similarity measures evaluate consistency between textual claims and actual metrics. Temporal features monitor consistency across multiple time periods, identifying trends of unfulfilled promises. Sentiment analysis components detect emotionally manipulative language that may mask negative performance indicators.

[0081] The system implements a Multi-Agent System (MAS) framework where each module / agent is responsible for discrete tasks, including ESG claim extraction, discrepancy detection, and greenwashing classification. The agents operate in a coordinated pipeline, with each stage contributing toward a robust, evidence-based evaluation of corporate sustainability claims.

[0082] The system significantly advances the art of automated ESG evaluation by introducing the jurisdiction (e.g., India) centric AVML-driven system for specific (e.g., BRSR) compliance and greenwashing detection. It empowers corporations with the tools to improve ESG transparency, provides investors with reliable information for value-aligned investment decisions, and sets a technological precedent for integrity in sustainability disclosures across sectors.

[0083] For better understanding, one or more embodiments of the present invention shall be described with respect to the earlier-mentioned drawings.

[0084] FIG. 1 illustrates a schematic diagram 100 of the document processing pipeline showing how raw ESG disclosure documents are processed into machine-readable formats through the retrieval and segmentation module, in accordance with an exemplary embodiment of the present disclosure. FIG. 2 illustrates a block diagram 200 of a vision-based OCR component within the retrieval and segmentation module for converting scanned ESG reports into structured text while preserving formatting and visual elements, in accordance with an exemplary embodiment of the present disclosure. FIG. 3 illustrates a schematic representation 300 of the retrieval and segmentation module organizing extracted content based on ESG principles Pl to P9 using a predefined taxonomy, in accordance with an exemplary embodiment of the present disclosure.

[0085] As illustrated, the implementation of the document ingestion and segmentation pipeline within a computer-implemented method for automated processing of ESG disclosure documents are described. These components support the broader objective of the invention: enabling high- accuracy, Al-driven, and modular ESG assessments for compliance, benchmarking, and greenwashing detection.

[0086] FIG. 1 illustrates a schematic representation of the document processing pipeline, which forms the initial stage of the automated processing of Environmental, Social and Governance (ESG) disclosure documents. The pipeline is implemented within a document processing agent 102, a core component of a retrieval and segmentation module. This agent is configured to receive raw ESG disclosure documents 104, including but not limited to BRSR (Business Responsibility and Sustainability Report) filings, typically in PDF format or as scanned image files.

[0087] Upon receiving such disclosure documents 104, the document processing agent 102 employs a Vision-Based Optical Character Recognition (OCR) module 112 to convert multipleinput images 108 into machine -readable text 118. This OCR module 112 processes each image 116 iteratively, recognizing and extracting textual content, while preserving structural elements such as headings, tables, and metadata. The extracted content 118 is then fed to a Markdown converter module 114, which transforms the content into a standardized Markdown format 106. This intermediate format retains the original visual hierarchy and formatting structure of the ESG disclosure document, facilitating downstream machine -based analysis and preserving contextual fidelity.

[0088] The output of this initial document processing operation is a structured, text-based machine-readable representation 106 of the original ESG disclosure. This enables downstream components of the system, including semantic search, evaluation, and validation modules, to operate efficiently and accurately on the processed content. The system is capable of handling high-volume document ingestion with minimal human intervention, thus enhancing scalability for enterprise-level ESG assessment applications.

[0089] FIG. 2 illustrates a block-level data flow diagram 200 of the Vision-Based OCR agent 202, which forms a subcomponent of the document processing agent 102 described in relation to FIG. 1. The OCR agent 202 corresponds to OCR module 112 is configured to process ESG PDFs or scanned disclosures 204 through a sequence of machine vision and natural language processing operations. First, the input document undergoes image processing 206 to enhance legibility, correct distortions, and optimize layout recognition. This stage includes noise removal, contrast enhancement, and page boundary detection to support more accurate text extraction.

[0090] The processed image data is then subjected to text recognition 208 using advanced OCR techniques that utilize machine learning-based character recognition models trained on domainspecific ESG report datasets. This step involves the identification of alphanumeric content, including technical and legal ESG terminology. Simultaneously, structure detection 210 is performed to identify visual elements such as tables, bullets, sections, and headings, ensuring the document's semantic structure is not lost. This is crucial for downstream segmentation and classification processes.

[0091] Finally, Markdown formatting 212 converts the output into a structured text document 214. The structured output 214 is semantically aligned with the original document layout, facilitating further segmentation and taxonomy-driven classification of ESG content by principle and category. The structured text 214 is consistent with the format 106 output by the document processing agent 102 in FIG. 1.

[0092] FIG. 3 presents a segmentation data flow diagram 300 implemented by a segmentationagent 302, another subcomponent of the retrieval and segmentation module. The segmentation agent 302 receives the machine-readable, structured text 106 generated from the OCR process and performs logical segmentation of content based on predefined ESG principles Pl through P9. These principles are derived from established frameworks such as the SEBI BRSR structure, Global Reporting Initiative (GRI), and other international ESG standards.

[0093] The segmentation agent 302 includes two key sub-agents: (i) a text segmentation agent 306, which divides the full document into discrete segments 310, each representing a logically independent unit (e.g., a disclosure statement, a table, or a policy excerpt); and (ii) a text classification agent 308, which further processes each segment 312 and assigns it one or more ESG-specific labels 304 based on the content type and relevance. These labels correspond to Environmental (E), Social (S), or Governance (G) categories and are also mapped to one of the predefined ESG principles P1-P9.

[0094] This dual- stage segmentation process enables precise targeting of content for ESG analysis, allowing the system to match relevant disclosures to specific evaluation criteria within the ESG frameworks. For example, a disclosure about carbon emissions might be labeled under Environmental (E) and matched with Principle P3 (Energy and Emissions), while a segment describing board diversity policies would fall under Governance (G), Principle P7 (Diversity and Inclusion).

[0095] In some embodiments, the segmentation module is enhanced through pattern recognition and named entity recognition techniques to detect quantitative ESG metrics (e.g., GHG emissions in tons CO2e, employee turnover percentages) and qualitative sustainability claims (e.g., commitment to net-zero). Segmented content is indexed and passed to subsequent modules for vectorization, semantic retrieval, and ESG evaluation.

[0096] The processing architecture shown in FIGS. 1-3 is implemented in a multi-agent system (MAS) as described in the present disclosure. Each agent namely, the Vision-Based OCR Agent, the Segmentation Agent, and downstream Analysis and Scoring Agents, functions autonomously but coordinates through a central orchestrator module, as described in subsequent part of this disclosure. The orchestrator module dynamically manages data flows between agents based on the document type and applicable ESG reporting framework, including BRSR, GRI, SASB, and TCFD.

[0097] In an exemplary use case, a scanned BRSR report from an Indian company is ingested into the system. The Vision-Based OCR agent 202 processes the document into structured text 214, preserving complex table structures and metadata. The Segmentation Agent 302 then classifies the extracted segments 310 under appropriate ESG categories and principles 304, enabling a rule-based evaluation engine to assess disclosures against benchmarks such as NGRBC, BIS, and SEBI mandates. The result is a validated and scored ESG output, compiled into a structured, human- readable report complete with visualizations.

[0098] FIG. 4 illustrates a process diagram 400 of the vectorization and storage module converting segmented ESG content into vector embeddings and storing them in a searchable vector database, in accordance with an exemplary embodiment of the present disclosure. FIG. 5 illustrates a functional diagram 500 of the analysis and validation module retrieving external benchmark data from at least one second source to verify ESG disclosures, in accordance with an exemplary embodiment of the present disclosure. FIG. 6 illustrates a block diagram 600 of the semantic retrieval module retrieving contextually relevant ESG content from the vector database using semantic search, in accordance with an exemplary embodiment of the present disclosure. FIG. 7 illustrates a functional view 700 of the analysis and validation module evaluating ESG disclosures against predefined ESG standards and detecting unverifiable claims, in accordance with an exemplary embodiment of the present disclosure. FIG. 8 illustrates a system diagram 800 of the scoring module assigning ESG scores to disclosure content using a 100-point rule-based or weighted scoring mechanism, in accordance with an exemplary embodiment of the present disclosure.

[0099] FIG. 4 illustrates a data flow diagram 400 of the vectorization and storage module, which converts segmented ESG content into vector embeddings and associated metadata, subsequently storing the combined data in a searchable vector database. In an exemplary embodiment, this function is performed by a Vectorization Agent 402 that receives segmented ESG content 404 (corresponding to output 304 of FIG. 3), wherein each segment is previously labeled with ESG categories (Environmental, Social, Governance) and ESG principles (P1P9). The Vectorization Agent 402 processes each text segment 416 and generates a vectorized representation 422 capturing the semantic context of the content using state-of-the-art embedding models, such as those derived from BERT, Sentence Transformers, or similar transformer-based natural language processing (NLP) architectures. Concurrently, metadata corresponding to principle classification 420 is generated and stored as labeled metadata 424.

[0100] The vectorized text and associated metadata are combined as enriched data 426 and stored across multiple database segments, including (i) current document data 438, (ii) ESG principlewise historical data 436 accumulated over time, (iii) identified industry regulations and guidelines 440 such as GRI, SASB, and TCFD, and (iv) identified industry best practices 442. The metadata generator 430, also part of the Vectorization Agent 402, processes instances of intermediate data 428 received from other system agents including the Web Research Agent 406, Analysis Agent408, Scoring Agent 410, Feedback Agent 412, and Report Generation Agent 414. These are converted into structured text data 432 and vectorized into embeddings 434 alongside their metadata, forming a composite ESG intelligence database that supports semantic search and contextual analysis.

[0101] FIG. 5 illustrates a data flow diagram 500 of the analysis and validation module, wherein the Web Research Agent 502 performs external benchmarking and validation of ESG disclosures. During process flow 504, the system generates one or more queries 518 based on ESG claims and disclosures identified within corporate reports. The Query Processor 510, a component of the Web Research Agent, refines and formats the queries 520 based on contextual understanding of ESG concepts and domain-specific terminologies. These processed queries are transmitted to a Data Fetcher and Updater 512, which retrieves up-to-date data from external web sources 506 such as GRI, SASB, UNPRB, and other international frameworks.

[0102] The retrieval and vectorization of external benchmark data is achieved via embedded Retrieval Agent 514 and Vectorization Agent 516, both internal to the Web Research Agent 502. The fetched benchmark data 508 is used to validate the company's ESG claims against industry standards and detect inconsistencies or unverifiable disclosures, thereby enhancing reliability and exposing potential greenwashing practices. The fetched data is stored and reused to enrich ESG evaluations across timeframes.

[0103] FIG. 6 presents a data flow diagram 600 of the Semantic Retrieval Module, which includes a Retrieval Agent 602 responsible for retrieving contextually relevant ESG content from the vector database. The Retrieval Agent receives a processed semantic query 604 (corresponding to query 518) and uses its Query Processor 608 to refine the query into an optimized form 622. This optimized query is submitted to the vector database, drawing upon segmented document data 610, industry regulations 612, historical ESG data 614, and best practices 616.

[0104] The Data Fetcher 618, a sub-component of the Retrieval Agent 602, identifies and returns highly relevant data 606 based on semantic similarity rather than keyword matching. This approach supports concept-level understanding and enables detection of substantively similar disclosures using differing terminology. The relevant content 606 is then returned to the analysis agents, enabling accurate context-aware assessments of ESG disclosures and their alignment with known frameworks.

[0105] The performance of the semantic retrieval module was evaluated against a keyword-based baseline using standard information retrieval metrics. As depicted in a table A hereinbelow, the semantic retrieval system demonstrated a Precision® 10 of 0.847, significantly outperforming thebaseline precision of 0.623, an improvement of 36.0%. This implies that 8.47 out of 10 retrieved ESG disclosure documents were relevant, compared to 6.23 in the baseline, thereby reducing the false positive rate from 37.7% to 15.3% and minimizing analyst review time for irrelevant content.Table - A

[0106] The Recall® 10 improved from 0.541 to 0.782 (a 44.5% increase), indicating the semantic retrieval captured a substantially greater portion of relevant ESG-specific content. This enhancement reduces the likelihood of missing critical ESG disclosures, thereby improving the completeness of compliance and validation processes.

[0107] In terms of overall ranking quality, the system achieved a Mean Average Precision (MAP) of 0.823, compared to 0.587 from the key word -based system, a 40.2% improvement. This demonstrates more consistent retrieval of relevant results across all recall levels, supporting efficient ESG due diligence.

[0108] The Normalized Discounted Cumulative Gain (NDCG@ 10) increased from 0.694 to 0.891 (28.4% gain), reflecting improved ranking of relevant disclosure documents at top positions. This reduces the number of search iterations needed for comprehensive ESG analysis, and enhances user experience during automated or human-assisted content review within the ESG report generation module.

[0109] FIG. 7 illustrates the functional operations 700 of the ESG Principle Analysis Agent 702 as part of the analysis and validation module. Each ESG content segment in the current document 704 is analyzed by an Analyzer component 716, which evaluates the content in relation to internal system agents and external validations. The analysis includes cross-referencing claims with web research data from the Web Research Agent 710, scoring data from the ESG Scoring Agent 712, and user / system feedback from Feedback Agent 714. Additionally, the Analysis Agent utilizes internal resources such as the Retrieval Agent 720 and the Vectorization Agent 718 to contextualize and verify claims / statements.

[0110] The Analyzer 716 generates an analyzed report 708 that reflects disclosure quality, specificity, completeness, and transparency. For example, the system may flag statements such as "We aim to reduce emissions" as vague unless accompanied by quantitative metrics and targettimelines. The analysis identifies missing documentation, evaluates alignment with global standards, and determines the accuracy of the sustainability assertions. This facilitates the detection of potentially misleading statements or unsubstantiated claims (i.e., greenwashing) and prompts recommended improvements.

[0111] FIG. 8 provides a data flow diagram 800 for the ESG Scoring Agent 802, which implements a structured scoring mechanism to quantitatively assess ESG disclosures. The agent receives the analyzed report 804 and uses the Score Generator 814 to produce an ESG score and explanatory remarks 806. This scoring process is informed by the Analysis Agent 808, Web Research Agent 810, and Feedback Agent 812, along with internal resources such as the Vectorization Agent 816 and Retrieval Agent 818.

[0112] The scoring framework is rule-based and / or weighted and evaluates disclosures using four primary criteria: (i) Key Strengths (20%) assesses exemplary practices, transparency, and third- party certifications; (ii) Improvement Suggestions (25%) identifies inconsistencies, framework misalignments, or incomplete disclosures; (iii) Specificity (25%) evaluates quantification, precision of metrics, target-setting, and geographic or temporal clarity; and (iv) Supporting Evidence (30%) validates claims through internal documentation and external verification. The resulting ESG score is on a 100-point scale, enabling consistent, comparative evaluations across sectors and time periods.

[0113] The reliability of the scoring module has been validated through expert evaluations to ensure consistency and credibility of assigned ESG scores. A comprehensive inter-rater reliability study was conducted involving 18 domain experts, including certified sustainability auditors, ESG analysts, and BRSR compliance specialists. The experts independently scored 120 ESG disclosure documents across six distinct industry sectors to assess scoring consistency.

[0114] The results demonstrated strong agreement among expert evaluators, with a Cronbach's Alpha of 0.88 indicating high internal consistency, an Intraclass Correlation Coefficient (ICC) of 0.82 reflecting good to excellent agreement, and a Fleiss' Kappa of 0.76 indicating substantial inter-rater reliability. To further validate alignment between Al-generated scores and expert judgment, the system achieved a Pearson correlation coefficient of r = 0.84, denoting strong positive correlation. The mean absolute error between Al and expert scores was 4.1 points on the 100-point scale, and the score variance (o2= 28.3) closely matched that of expert raters (o2= 31.7). Notably, 79% of Al-generated ESG scores fell within ±5 points of the expert consensus, and 93% were within ±10 points, confirming high agreement levels.

[0115] Scoring consistency was further validated across individual scoring dimensions definedin the weighted evaluation scheme. Correlation between expert and Al scores for key strengths (20%) was 0.81 (inter-expert ICC: 0.79); for improvement areas (25%), 0.86 (ICC: 0.83); for specificity (25%), 0.89 (ICC: 0.85); and for supporting evidence (30%), 0.82 (ICC: 0.80). The overall ESG score yielded an expert-AI correlation of 0.84, with an inter-expert ICC of 0.82, demonstrating the scoring module's ability to replicate expert-level evaluations with statistically validated consistency.

[0116] This modular and hierarchical ESG evaluation system thus integrates the Vectorization Agent, Web Research Agent, Semantic Retrieval Module, ESG Principle Analysis Agent, and ESG Scoring Agent into an end-to-end pipeline. The architecture supports dynamic, real-time analysis and can be adapted to evolving ESG standards and reporting obligations. The system is particularly suited for automated processing of BRSR reports submitted by public companies and may be extended to any structured or semi-structured sustainability disclosures.

[0117] FIG. 9 illustrates a schematic diagram 900 of the orchestrator module dynamically coordinating inter-module data flows and applying framework- specific processing logic, in accordance with an exemplary embodiment of the present disclosure. FIG. 10 illustrates a process diagram 1000 of the report generation module synthesizing ESG scoring and validation results into structured, human-readable ESG reports with visual outputs, in accordance with an exemplary embodiment of the present disclosure. FIG. 11 illustrates a system workflow diagram 1100 showing interaction among all modules from document ingestion to ESG insight generation coordinated by the orchestrator module, in accordance with an exemplary embodiment of the present disclosure. FIG. 12 illustrates an architectural overview 1200 of the system comprising six processing modules arranged around the orchestrator module in a hub-and-spoke configuration, in accordance with an exemplary embodiment of the present disclosure.

[0118] FIG. 9 illustrates a schematic diagram 900 of the Feedback Agent 902, corresponding to 412 of Fig. 4, functioning as the orchestrator module. This module is responsible for dynamically coordinating inter-module data flows and applying framework specific processing logic. The Feedback Agent 902 receives an analyzed report 904 (corresponds to 708 and scoring data with remarks 906 (corresponds to 806 of Fig. 8) from various analytical agents. It generates a synthesized feedback report 908 via its internal feedback generator 916. This generation is based on inputs from the Analysis Agent 910 (corresponds to 408 of Fig. 4), Scoring Agent 912 (corresponds to 410 of Fig. 4), and Web Research Agent 914 (corresponds to 406 of Fig. 4), which are external to the orchestrator, as well as the internal Vectorization Agent 918 (corresponds to 402 of Fig. 4) and Retrieval Agent 920 (corresponds to 514 of Fig. 5).

[0119] Strategically, the Feedback Agent 902 functions as a supervisory wrapper layer thatcoordinates the flow of information across the system. Bidirectional green pathways shown in the diagram depict dynamic communication channels. These facilitate real-time validation, quality assurance, and methodological consistency in ESG evaluation across diverse reports. Moreover, connections to external components such as "Annotated Papers" and the "Input and Access Dashboard" enable data enrichment and user interactivity.

[0120] This configuration ensures consistent benchmarking, cross-validation, and continuous error correction. The orchestrator module identifies discrepancies, mitigates false positives, and enables adaptive learning to detect nuanced greenwashing strategies. The system thereby enhances sustainability assessment fidelity through supervised, iterative refinement.

[0121] FIG. 10 presents a detailed process diagram 1000 of the Report Generation Agent 1002, corresponding to 414 of Fig. 4. This module synthesizes ESG scoring and validation outputs into structured, human-readable reports enriched with visual elements. The agent receives three main inputs: analyzed report 1004 (corresponds to 708 of Fig. 7), feedback report 1006 (corresponds to 908 of Fig. 9), and scoring data 1008 (corresponds to 806 of Fig. 8). Internally, the PDF generator 1020 compiles these inputs into the final generated PDF report 1010. This transformation is supported by external agents, including Analysis Agent 1012 (corresponds to 408 of Fig. 4), Feedback Agent 1014 (corresponds to 412 of Fig. 4), Scoring Agent 1016 (corresponds to 410 of Fig. 4), and Web Research Agent 1018 (corresponds to 406 of Fig. 4), and internal modules such as Vectorization Agent 1022 (corresponds to 402 of Fig. 4), Retrieval Agent 1024 (corresponds to 514 of Fig. 5), and an Evaluator 1026.

[0122] Functionally, the Report Generation Agent serves as the system's synthesis hub. It aggregates multi-source outputs and formats them into standardized reports. Each report includes an executive summary, radar charts, benchmark comparisons, trend analysis graphs, and color- coded heatmaps. Additionally, principle-wise breakdowns with supporting evidence, greenwashing risks, and actionable recommendations are included. Appendices provide methodology references, compliance status, and historical ESG trends.

[0123] Bidirectional purple arrows in coordination with Evaluator 1026, in the diagram denote continuous feedback loops that facilitate iterative report refinement. Integration with the Document Object Model (Structured ESG Report) ensures adherence to established reporting frameworks and consistency in output formatting. Reports generated contribute to methodological optimization through downstream interaction with the Methodological Agent and Retrieval Agent.

[0124] FIG. 11 illustrates an end-to-end workflow diagram 1100 representing systemwide interactions from document ingestion to ESG insight generation. The flow begins with theDocument Processing Agent 102, which transforms the BRSR report 104 into structured markdown format text 106.

[0125] Subsequently, the Segmentation Agent 302 classifies the text 106 into segments labeled by ESG principles P1-P9 and categorizes them into Environmental, Social, or Governance buckets 304. These segments are then passed to the Vectorization Agent 402, which transforms the content into semantic embeddings and augments it with:• Historical ESG data 436,• Current report information 438,• Industry regulations and guidelines 440,• Best practices 442.

[0126] Following vectorization, the data is evaluated by the ESG Principle Analysis Agent 702, composed of the Analysis Agent 408, Scoring Agent 410, and Feedback Agent 412. External validation is performed by the Web Research Agent 406 using web data 506. The consolidated output is passed to the Report Generation Agent 414, which generates the final structured ESG report 1010. This workflow ensures comprehensive cross-checking, scoring, and validation across modules, resulting in an accurate, holistic ESG profile for each company.

[0127] FIG. 12 presents an architectural block diagram 1200 illustrating the full system in a hub- and-spoke arrangement. The central hub consists of the Feedback Agent 412, which orchestrates the operation and communication among six primary modules:• Document Processing Agent 102 - Performs OCR and text extraction from BRSR documents.• Segmentation Agent 302 - Organizes extracted text into ESG principle -based segments.• Web Research Agent 406 - Conducts external validation against industry benchmarks and regulatory data.• ESG Principle Analysis Agent 702 - Evaluates credibility and specificity of disclosures.• Scoring Agent 410 - Applies a 100-point structured evaluation methodology with weighted criteria.• Report Generation Agent 414 - Synthesizes all outputs into structured PDF reports.

[0128] This modular configuration allows scalable, maintainable integration of advanced Al technologies, including:• LangChain for natural language processing,• Hugging Face Transformers for deep text analysis,FAISS for vector-based semantic search.

[0129] These tools empower the system to efficiently process large volumes of ESG content, apply rigorous evaluations, and generate high-fidelity reports. The feedback loop enabled by the Feedback Agent ensures error correction, cross-validation, and ongoing learning, adapting the system to evolving ESG reporting practices and emerging greenwashing techniques.

[0130] For example, when processing a BRSR report from a leading Indian manufacturing company, the system extracts sustainability disclosures using the Document Processing Agent. The Segmentation Agent identifies principle- specific disclosures (e.g., waste management under P2). The Vectorization Agent encodes this information, and the ESG Principle Analysis Agent scores the claim for specificity, accuracy, and alignment with known benchmarks. The Web Research Agent verifies the existence of relevant third-party certifications. Scores and remarks are passed to the Feedback Agent, which coordinates consistency checks across agents. The final report generated includes comparative radar charts, compliance heatmaps, and industry benchmarks, offering both qualitative and quantitative insight.

[0131] This end-to-end system thus ensures accurate, reproducible, and comprehensive ESG assessments while reducing the risk of subjective bias and greenwashing. It provides stakeholders including companies, investors, and regulators, with actionable intelligence grounded in rigorous methodological evaluation.

[0132] FIG. 13 illustrates a block diagram 1300 of system hardware comprising processor, memory, communication interface, and functional modules including retrieval and segmentation module, vectorization and storage module, semantic retrieval module, analysis and validation module, scoring module, and orchestrator module, in accordance with an exemplary embodiment of the present disclosure.

[0133] FIG. 13 provides a holistic overview of the computational infrastructure enabling the intelligent ESG evaluation and greenwashing detection framework. The described system is scalable, modular, and interoperable, designed to handle high volumes of unstructured data and convert them into actionable sustainability insights using state-of-the-art artificial intelligence techniques. The computational infrastructure is configured to implement the ESG evaluation and greenwashing detection framework as described in previous embodiments.

[0134] The system includes a combination of hardware components and software modules interconnected via communication channels and designed to execute a plurality of functionalities in accordance with the objectives of the present disclosure. As shown, the system hardware includes at least one processor 1302, one or more memory units 1308, a communication interface1304, and a communication channel 1306, operatively coupled to enable seamless intercomponent operation. These components interact with a plurality of functional modules, each responsible for a discrete stage in the ESG assessment and report generation pipeline.

[0135] The processor 1302 may be implemented as one or more central processing units (CPUs), microcontrollers, digital signal processors (DSPs), logic circuits, microprocessors, or other processing elements capable of executing instructions. The processor is configured to execute computer-readable instructions 1312 stored in memory 1308 to perform operations such as text extraction, semantic vectorization, retrieval, scoring, and report synthesis. In certain embodiments, the processing unit may be part of a distributed architecture that supports parallel processing of multiple ESG documents.

[0136] The memory 1308 is used for storing both data 1310 and executable instructions 1312. The memory may include volatile memory such as RAM, as well as non-volatile memory including ROM, EEPROM, flash memory, hard drives, or solid-state drives. The stored instructions implement the functionalities of the modules described herein, including logic for document parsing, agent orchestration, benchmarking, validation, and output formatting. Memory also supports intermediate storage of tokenized text, scoring outputs, and formatted reports.

[0137] The communication interface 1304 enables the system to interact with external devices, databases, cloud infrastructure, and user dashboards via wired or wireless networks. The network interface may support Ethernet, USB, Wi-Fi, Zigbee, Bluetooth, or cellular technologies including 2G, 3G, 4G, LTE, 5G, or 6G. This allows the system to fetch external regulatory documents, benchmark datasets, or update model weights from distributed servers in real-time.

[0138] In one embodiment, the communication channel 1306 interconnects all modules, allowing asynchronous data sharing and module- specific orchestration across the hardware. This ensures optimal bandwidth utilization and synchronized operation of various data pipelines.

[0139] The hardware further integrates the specialized functional modules:• Retrieval and Segmentation Module 1314: Responsible for extracting machine-readable text from ESG reports and segmenting it into logical units according to the nine ESG principles (P1-P9). This module may employ OCR for PDF extraction and named-entity recognition for contextual segmentation.• Vectorization and Storage Module 1316: Converts the segmented text into semantic embeddings using natural language processing models such as BERT or SentenceBERT. The resulting vector representations are stored in a semantic database (e.g., FAISS) for fast retrieval and similarity search.• Semantic Retrieval Module 1318: Enables efficient retrieval of contextually relevant data across current and historical ESG disclosures. This module utilizes cosine similarity or nearest neighbor algorithms to fetch embeddings related to a given principle, claim, or keyword.• Analysis and Validation Module 1320: Implements Al -based validation mechanisms, including claim- specific credibility assessment and greenwashing detection. It integrates outputs from various agents (e.g., Web Research Agent, Analysis Agent) and applies cross- validation against external data sources and historical trends.• Scoring Module 1322: Assigns structured, weighted ESG scores based on predefined criteria. This module processes inputs from the analysis and validation layer and computes principle-wise and aggregated ESG scores, which feed into the final report.• Orchestrator Module 1324: Serves as the supervisory coordination engine managing intermodule communication and workflow optimization. It dynamically triggers modules, handles exception flows, and ensures consistency in ESG evaluation methodologies.

[0140] The system architecture further supports interaction with external computers 1326, enabling real-time communication with user terminals, regulatory databases, corporate data repositories, or cloud-based Al services. These external systems may be utilized for web scraping, dynamic benchmarking, regulatory updates, and cloud storage of generated ESG reports.

[0141] In certain embodiments, the system includes one or more VO Managers, implemented in software or firmware, capable of supporting multiple communication protocols such as MQTT, OPC UA, Modbus, SECS / GEM, and Profinet. These allow for integration with various industrial or enterprise systems, thereby expanding the utility of the ESG evaluation framework to a wider range of data environments.

[0142] Additionally, the system's architecture supports extensibility for future enhancements such as the inclusion of domain- specific ESG indicators, real-time greenwashing risk alerts, integration with loT-based environmental sensors, and adaptive Al retraining capabilities, thereby ensuring compliance with evolving standards in sustainability disclosures.

[0143] Empirical benchmarking of the end-to-end ESG processing pipeline coordinated by the orchestrator module demonstrates significant efficiency and scalability across varying document sizes and batch workloads. The average processing time per ESG disclosure document is approximately 95 seconds, with smaller BRSR reports (10-25 pages) processed in 45-60 seconds, medium- length reports (26-50 pages) in 85-120 seconds, and larger reports (51-100+ pages) in 180-240 seconds. The processing workload is distributed across stages, with OCR and text extraction accounting for 25-30%, segmentation and classification for 15-20%, vectorization for20-25%, analysis and validation for 25-30%, and report generation for 10-15% of the total time.

[0144] Parallel execution capability enables the system to process 1,000 documents in approximately 15.2 hours (averaging 55 seconds per document), scaling to 2,500 documents in 34.8 hours. Efficiency gains through parallelization demonstrate a 9. lx speedup using 16 parallel workers, with diminishing returns observed beyond 32 workers. The system's memory usage per document includes base memory of 2.1 GB, peak memory of 850 MB to 1.2 GB, vector database storage of 15-25 MB, and temporary processing file size of 5-8 MB. For batch processing, memory usage ranges from 12-16 GB for 100 documents to 45-60 GB for 1,000 documents, with a concurrency cap of 32 documents on a 64 GB system.

[0145] The vector database, implemented using FAISS with IVF-PQ indexing, grows proportionally with document count - occupying 18.5 GB for 1,000 documents and 167.8 GB for 10,000 documents, with an effective compression ratio of 3.2: 1 compared to raw text. Semantic search latency ranges from 15-45 milliseconds, with full index rebuilds requiring 2.3 hours for 10,000 documents, and backup / restore operations completed within 23 minutes.

[0146] The system's robustness and real-world scalability were demonstrated through large-scale deployments such as the 2023 analysis of BRSR reports from NIFTY 500 companies, where 2,847 documents were processed in 72 hours, achieving an average processing time of 91 seconds / document and system availability of 99.7% with zero data loss. A broader validation study covering 15,000 ESG reports over five years confirmed stable throughput with 78% CPU and 85% memory utilization.

[0147] Compared to manual ESG analysis by human experts-which typically requires 4-6 hours per document, yields 73% inter-rater agreement, costs ?2, 500-4,000 per document, and achieves a throughput of 2-3 documents per analyst per day-the automated system processes documents in 95 seconds, reaches 97% consistency (validated against expert consensus), costs ?12.50 per analysis, and supports a throughput of 850-950 documents per day per server. This results in 152x speed improvement, 99.5% cost reduction, 24% higher scoring consistency, and 300x greater throughput relative to manual ESG analysis workflows.

[0148] FIG. 14 illustrates a method flow diagram 1400 of the computer-implemented method for automated processing of ESG disclosure documents performed by the processor using the functional modules, in accordance with an exemplary embodiment of the present disclosure.

[0149] As illustrated, the method for the automated processing of ESG disclosure documents is described. The method is executed by a processor in conjunction with a plurality of functional modules, each performing discrete, logically ordered steps to facilitate machine-driven ESGassessment, scoring, and report generation.

[0150] At step 1402, the method begins with the segmenting, by the processor using a retrieval and segmentation module, of ESG-specific content sections from one or more ESG disclosure documents. These documents are retrieved from at least one first source, which may include enterprise data repositories, public databases, or regulatory filings. The segmentation process preserves structural document elements, such as section headings, tables, footnotes, and embedded metadata. The segmentation is performed in alignment with a predefined taxonomy of ESG principles, including Principles Pl to P9, to ensure thematic consistency and contextual traceability across disclosures.

[0151] At step 1404, the segmented content is converted, by the processor using a vectorization and storage module, into semantic vector representations using deep learning based language models, such as Sentence-BERT or domain- adapted transformers. The vectorized content is stored in a searchable vector database, implemented via a high performance similarity search engine, such as Facebook Al Similarity Search (FAISS). This enables efficient, low-latency retrieval operations for downstream semantic search.

[0152] At step 1406, the method proceeds to performing semantic search over the vector database using a semantic retrieval module, based on ESG-specific queries. The queries may be predefined or dynamically generated based on the reporting framework or industry sector of the disclosing entity. The semantic retrieval operation retrieves contextually relevant disclosure fragments and ranks the results based on semantic similarity. It also identifies and associates supporting evidence for each retrieved ESG content item.

[0153] At step 1408, the processor performs evaluation and validation of the retrieved disclosure content using an analysis and validation module. The evaluation involves comparing the content against predefined ESG standards and industry-specific benchmarks, including but not limited to the Global Reporting Initiative (GRI), Sustainability Accounting Standards Board (SASB), and the Task Force on Climate-related Financial Disclosures (TCFD). Validation further includes the comparison of internal claims against external benchmark data from at least one second source, such as regulatory filings, market intelligence platforms, or sustainability databases. The module is also configured to detect vague or unverifiable claims / statements indicative of greenwashing, based on semantic inconsistencies or unsupported assertions in the disclosure text.

[0154] At step 1410, the method advances to assigning ESG scores using a scoring module. The scoring is based on a rule -based or weighted evaluation mechanism, and in one embodiment, employs a 100-point scoring rubric comprising four weighted criteria: (i) Key Strengths, (ii)Improvement Suggestions, (iii) Specificity, and (iv) Supporting Evidence. The ESG scores may be assigned at the principle level (P1-P9) and aggregated to provide an overall ESG rating.

[0155] At step 1412, the processor proceeds with generating a structured ESG report using a report generation module. The report includes detailed performance metrics, scoring summaries, and visual outputs such as radar charts, heat maps, and benchmark comparisons. In one embodiment, the report generation module interfaces with a dashboard to support interactive user engagement, enabling stakeholders to explore principle-level details, download reports, or provide feedback for iterative improvements.

[0156] Finally, at step 1414, the processor employs an orchestrator module to dynamically coordinate the steps from segmenting to report generation. This orchestration manages intermodule data flows and adapts execution paths based on the content type and ESG reporting framework detected in the input documents. The orchestrator module maintains a framework identification knowledge base, which is continually updated using pattern recognition models trained on structural and semantic traits of various ESG standards. The orchestrator module also applies framework- specific error detection thresholds:• For GRI, it enforces section completeness validation;• For SASB, it applies industry-specific disclosure requirements;• For TCFD, it verifies inclusion of forward-looking climate scenario analyses.

[0157] Additionally, the orchestrator module supports version control by tracking framework evolution and applying version-specific processing rules, determined via document metadata. In cases where the disclosure is multi-framework in nature, the orchestrator module implements hybrid processing paths, detects overlapping framework requirements, and resolves conflicting expectations using materiality-based prioritization techniques.

[0158] The method described above provides a comprehensive, modular, and adaptive workflow for automated ESG document processing, enabling robust greenwashing detection, regulatory compliance, and transparency in sustainability disclosures. The integration of Albased semantic processing, benchmarking, and dynamic orchestration renders the system scalable and applicable across industries, regions, and ESG reporting standards.Table B

[0159] As outlined in the above table - B, the greenwashing detection functionality of the analysis and validation module was experimentally validated using a curated dataset of 2,850 ESG disclosure documents annotated by qualified ESG professionals. The annotation process achieved substantial inter-rater agreement, as indicated by a Fleiss' Kappa of 0.82. On this dataset, the system achieved a precision of 0.891 (95% CI: 0.873-0.908), recall of 0.927 (95% CI: 0.912-0.941), and Fl-score of 0.909 (95% CI: 0.896-0.921). The system's overall greenwashing detection accuracy was 0.934, and the model statistically outperformed existing baseline approaches (p < 0.001) across multiple industries and reporting periods.

[0160] In a comparative benchmark study, the greenwashing detection capabilities of the present system were measured against those of leading ESG scoring providers and manual expert analysis. The Al-powered system achieved a 91% detection rate, outperforming Refinitiv ESG (75%), Bloomberg ESG (71%), and MSCI ESG Research (79%) - a +12 percentage point improvement over the best alternative. The system also recorded the lowest false positive rate at 8%, compared to 16-28% for other methods.

[0161] The Al system further achieved a 97% correlation with expert consensus in ESG scoring accuracy, compared to 73% for manual analysis and 84% for MSCI. It demonstrated 94% accuracy in ESG principle classification across P1-P9 and offered substantial operational advantages - including 95-second processing time per document, 950-document daily throughput per server, and daily update cycles, all of which exceeded the performance benchmarks of existing solutions. Additionally, the system provides 100% automated audit trails, full ML-based version control, and support for 25+ regulatory frameworks, including the BRSR, establishing it as a state-of-the-art solution for ESG disclosure validation and greenwashing detection.

[0162] The following are a few of the major advantages of the present invention:• The present invention provides a unified system and method that combines AVMLbased analytics, disclosure-performance mapping, and regulatory compliance verification for comprehensive ESG assessment, including alignment with India's BRSR framework.• The present invention enables automated extraction and evaluation of ESG disclosures from structured and unstructured sources, with specific capabilities to detectmisleading ’’green language,” vague generalizations, and linguistic markers indicative of greenwashing.• The present invention utilizes unsupervised learning algorithms to uncover hidden anomalies, data inconsistencies, and outliers in ESG content, flagging deceptive disclosure behaviors such as selective reporting and exaggerated sustainability claims.• The present invention benchmarks ESG content against industry-specific standards and generates projected ESG scores. Provides actionable feedback including:(a) Identification of greenwashed content, (b) Tailored improvement suggestions, and(c) Classification into Laggard, Average, or Leader categories• The present invention ensures all ESG evaluation parameters are aligned with India's mandatory BRSR (Business Responsibility and Sustainability Reporting) framework, covering all nine principles such as ethics, human rights, environmental conservation, and consumer well-being.• The present invention assesses ESG disclosures for compliance with multiple reporting standards and regulatory requirements including SEBI, GRI, SASB, and BRSR, improving regulatory alignment and audit-readiness.• The present invention offers automated linguistic and semantic feedback to end-users for iterative refinement of ESG content. Includes real-time visualizations of greenwashing probability, content-level scoring, and performance heatmaps.• The present invention facilitates intelligent parsing and data extraction from cloudbased or distributed data repositories, allowing scalable and standardized report generation across diverse sectors and regulatory regimes.• The present invention equips investors, regulators, and stakeholders with credible and benchmarked ESG insights, reducing information asymmetry and mitigating risks associated with unverifiable sustainability claims.• The present invention provides contextual performance evaluation by comparing an entity's ESG disclosures against industry peers. Highlights ESG maturity gaps and best practices based on sector-specific disclosure trends.• The present invention implements a multi-tiered scoring system tailored to sectoral ESG practices, classifying firms into Laggard, Average, and Leader categories to indicate relative ESG maturity and compliance levels.• The present invention employs Al models trained on sectoral historical datasets to identify greenwashing trends unique to industries such as manufacturing, energy, FMCG, and IT, enabling targeted risk assessment.• The present invention supports explain ability in Al-based ESG analysis by identifying andquantifying inconsistencies between stated disclosures and actual operational sustainability metrics, enhancing transparency and accountability.• The present invention generates final ESG reports that present both a projected ESG score and a quantified greenwashing score, thereby promoting integrity in ESG communication and enabling strategic alignment with real-world practices.• The present invention offers a highly modular and scalable ESG evaluation and greenwashing detection solution, suitable for integration and deployment by corporates, rating agencies, investors, and regulatory bodies across multiple geographies and sectors.

[0163] It is understood that any specific order or hierarchy of blocks in the processes disclosed is an illustration of example approaches. Based upon design preferences, it is understood that the specific order or hierarchy of blocks in the processes may be rearranged, or that all illustrated blocks be performed. Any of the blocks may be performed simultaneously. In one or more implementations, multitasking and parallel processing may be advantageous.

[0164] It should be noted that the description and figures merely illustrate the principles of the present subject matter. It should be appreciated by those skilled in the art that conception and specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present subject matter. It should also be appreciated by those skilled in the art by devising various systems that, although not explicitly described or shown herein, embody the principles of the present subject matter and are included within its spirit and scope.

[0165] Furthermore, all examples recited herein are principally intended expressly to be for pedagogical purposes to aid the reader in understanding the principles of the present subject matter and the concepts contributed by the inventor(s) to further the art and are to be construed as being without limitation to such specifically recited examples and conditions. The novel features which are believed to be characteristic of the present subject matter, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures.

[0166] Although embodiments for the present subject matter have been described in language specific to package features, it is to be understood that the present subject matter is not necessarily limited to the specific features described. Rather, the specific features and methods are disclosed as embodiments for the present subject matter. Numerous modifications and adaptations of the system / device of the present invention will be apparent to those skilled in the art, and thus it is intended by the appended claims to cover all such modifications and adaptations which fall within the scope of the present subject matter.

Claims

WE CLAIM:

1. A computer-implemented method for automated processing of Environmental, Social and Governance (ESG) disclosure documents, comprising the steps of:(a) segmenting (1402), by a processor (1302) using a retrieval and segmentation module, ESG- specific content sections from one or more ESG disclosure documents retrieved, by the processor using the retrieval and segmentation module, from at least one first source;(b) converting (1404), by the processor (1302) using a vectorization and storage module, the segmented content into vector representations and storing the vectors in a searchable vector database;(c) performing (1406), by the processor (1302) using a semantic retrieval module, semantic search over the vector database based on ESG- specific queries to retrieve contextually relevant disclosure content;(d) evaluating (1408), by the processor (1302) using an analysis and validation module, the retrieved content against predefined ESG standards and validating the authenticity based on external benchmark data obtained, by the processor using an analysis and validation module, from at least one second source;(e) assigning (1410), by the processor (1302) using a scoring module, ESG scores to the disclosure documents based on the evaluation and validation results based on a rulebased or weighted scoring mechanism;(f) generating (1412), by the processor (1302) using a report generation module, a structured, human-readable ESG report comprising ESG performance metrics and visual outputs, based on the scoring and validation results; and(g) dynamically coordinating (1414), by the processor (1302) using an orchestrator module, the sequential steps (a) to (f), the orchestrator being configured to manage inter-module data flows based on content type of the disclosure documents or ESG reporting framework.

2. The method as claimed in claim 1, wherein the scoring module applies a 100-point evaluation comprising four weighted criteria: key strengths, improvement suggestions, specificity, and supporting evidence.

3. The method as claimed in claim 1, wherein the analysis and validation module dynamically formulates queries based on the sector and operational domain of the disclosing entity.

4. The method as claimed in claim 1, wherein the analysis and validation module detects vague or unverifiable claims indicative of greenwashing, based on mismatches between internal disclosures and external data.

5. The method as claimed in claim 1, wherein the retrieval and segmentation module preserves structural elements such as headings, tables, and metadata during extraction.

6. The method as claimed in claim 1, wherein segmentation is performed based on a predefined taxonomy of ESG principles comprising principles Pl to P9.

7. The method as claimed in claim 1, wherein vector representations are computed using deep learning-based models and stored in the searchable vector database implemented based on a similarity search engine comprising Facebook Al Similarity Search (FAISS).

8. The method as claimed in claim 1, wherein the step of retrieving comprising ranking results based on contextual relevance and identifying supporting evidence for each ESG content.

9. The method as claimed in claim 1, wherein the evaluation is conducted based on industryspecific benchmarks, wherein the industry-specific benchmarks comprise the Global Reporting Initiative (GRI), Sustainability Accounting Standards Board (SASB), and the Task Force on Climate -related Financial Disclosures (TCFD).

10. The method as claimed in claim 1, wherein the report generation module produces the visual outputs comprising radar charts, heat maps, and benchmark comparisons, and wherein the report generation module interfaces with a dashboard to enable user-driven interaction and feedback incorporation.

11. The method as claimed in claim 1, wherein the step of dynamically coordinating, by the processor using the orchestrator module, comprises:(a) maintaining a framework identification knowledge base that is continually updated with pattern recognition models trained on structural and semantic characteristics of each ESG reporting framework;(b) applying framework- specific error detection thresholds, wherein the Global Reporting Initiative (GRI) framework processing implements section-completeness validation, the Sustainability Accounting Standards Board (SASB) framework processing enforces industry-specific disclosure requirements, and the Task Force on Climate -related Financial Disclosures (TCFD) framework processing verifies the inclusion of forwardlooking climate scenario analyses;(c) tracking framework evolution by maintaining version-specific processing rules for each ESG standard and automatically identifying the applicable framework version based on document metadata; and(d) implementing hybrid processing paths for multi-framework disclosures by detectingoverlapping framework requirements and reconciling conflicting disclosure expectations through materiality-based prioritization techniques.

12. A system for automated processing of Environmental, Social and Governance (ESG) disclosure documents, comprising:(a) a retrieval and segmentation module (1314) configured to retrieve one or more ESG disclosure documents from at least one first source, and segment ESG-specific content sections from the retrieved disclosure documents;(b) a vectorization and storage module (1316) configured to generate vector representations of the segmented content and store them in a searchable vector database;(c) a semantic retrieval module (1318) configured to retrieve contextually relevant disclosure content based on ESG-specific queries from the vector database;(d) an analysis and validation module (1320) configured to assess retrieved content against predefined ESG standards and validate the authenticity based on external benchmark data obtained from at least one second source;(e) a scoring module (1322) configured to assign ESG scores based on the evaluation and validation results using a rule -based or weighted scoring mechanism;(f) a report generation module configured to produce structured, human-readable reports comprising ESG performance metrics and visual outputs; and(g) an orchestrator module (1324) configured to coordinate the sequential operations of the modules (a) to (f), and to handle inter-module data flows dynamically based on content type of the disclosure documents or ESG reporting framework.

Images