Ganesh Harke
Research Subjects
Ganesh Harke has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
A. Panko
Independent
1 shared publication- Devdas Gupta1 shared publication
- Dipankar Saha
Independent Researcher
1 shared publication J. Joubert
Independent
1 shared publicationK. Hoo
Independent
1 shared publication- Lav Kumar1 shared publication
Latest Publications
- 7 January 2027
- Published by IGI Global Scientific Publishing
- p. 33-78
Engineering Human-in-the-Loop Control for Agentic AI
The transition from Predictive to Agentic AI transforms investment management systems from linear decision pipelines into recursive orchestrators, creating a governance deficit at the intersection of Models, Markets, and Mandates. Unlike passive predictive models, autonomous agents possess actionable intelligence that challenges the non-delegable nature of fiduciary duty. This chapter proposes an engineering-led framework to bridge this accountability gap by replacing manual oversight with a rigorous Control Plane architecture. It develops a taxonomy of Human-in-the-Loop designs, including the Centaur (Signal Generation), Governor (Execution), and Risk Crew (Incident Response) models, mapped to specific latency and risk profiles. By synthesizing recent engineering patterns such as defense prompts and state-aware kill switches, the chapter shows how to impose deterministic constraints on probabilistic planners. Finally, it reframes Chain-of-Thought logging as a structural mechanism for process auditability, ensuring that autonomous investment decisions meet the regulatory standards.
- 7 January 2027
- Published by IGI Global Scientific Publishing
- p. 33-78
Engineering Human-in-the-Loop Control for Agentic AI
The transition from Predictive to Agentic AI transforms investment management systems from linear decision pipelines into recursive orchestrators, creating a governance deficit at the intersection of Models, Markets, and Mandates. Unlike passive predictive models, autonomous agents possess actionable intelligence that challenges the non-delegable nature of fiduciary duty. This chapter proposes an engineering-led framework to bridge this accountability gap by replacing manual oversight with a rigorous Control Plane architecture. It develops a taxonomy of Human-in-the-Loop designs, including the Centaur (Signal Generation), Governor (Execution), and Risk Crew (Incident Response) models, mapped to specific latency and risk profiles. By synthesizing recent engineering patterns such as defense prompts and state-aware kill switches, the chapter shows how to impose deterministic constraints on probabilistic planners. Finally, it reframes Chain-of-Thought logging as a structural mechanism for process auditability, ensuring that autonomous investment decisions meet the regulatory standards.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
GuardLoop: A Runtime Governance Architecture for Human-Supervised Agentic AI
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
GuardLoop: A Runtime Governance Architecture for Human-Supervised Agentic AI
- 24 July 2026
- Published by IGI Global Scientific Publishing
- p. 113-154
Financial Data Engineering
Quantitative trading and risk management systems depend critically on the integrity of their underlying data infrastructure. While much of the academic literature treats financial data as static, preprocessed time series, real-world market data arrives as noisy, asynchronous streams requiring rigorous engineering controls. This chapter presents a comprehensive framework for Financial Data Engineering, tracing the lifecycle from raw market feeds to a point-in-time consistent Feature Store. It examines architectural patterns for ingestion, storage, transformation, and serving, emphasizing DataOps principles, polyglot persistence, and reproducibility across research and production environments. The chapter further addresses microstructure-aware cleaning, reconciliation, auditability, and synthetic stress-testing for rare market conditions. By formalizing market environments and standardizing feature computation, the proposed approach reduces look-ahead bias and mitigates the Research–Production Gap, providing a blueprint for robust, end-to-end financial data systems.
- 24 July 2026
- Published by IGI Global Scientific Publishing
- p. 113-154
Financial Data Engineering
Quantitative trading and risk management systems depend critically on the integrity of their underlying data infrastructure. While much of the academic literature treats financial data as static, preprocessed time series, real-world market data arrives as noisy, asynchronous streams requiring rigorous engineering controls. This chapter presents a comprehensive framework for Financial Data Engineering, tracing the lifecycle from raw market feeds to a point-in-time consistent Feature Store. It examines architectural patterns for ingestion, storage, transformation, and serving, emphasizing DataOps principles, polyglot persistence, and reproducibility across research and production environments. The chapter further addresses microstructure-aware cleaning, reconciliation, auditability, and synthetic stress-testing for rare market conditions. By formalizing market environments and standardizing feature computation, the proposed approach reduces look-ahead bias and mitigates the Research–Production Gap, providing a blueprint for robust, end-to-end financial data systems.
- 1 July 2026
- Published by IGI Global Scientific Publishing
- p. 167-202
Algorithmic Accountability in Greenwashing Detection
The rapid growth of ESG investing has intensified corporate greenwashing and overwhelmed traditional human auditing mechanisms. This chapter examines the comparative efficacy of Large Language Models and human analysts in detecting corporate greenwashing. As sustainable finance expands, deceptive environmental reporting exploits the capacity constraints, biases, and structural limitations inherent in manual auditing. To address these vulnerabilities, institutions deploy automated semantic analysis to systematically cross-reference corporate narratives against objective data. However, relying on opaque algorithms introduces novel risks, particularly statistical hallucinations and algorithmic audit washing, wherein flawed environmental claims receive unwarranted mathematical validation. The chapter advances a Human-in-the-Loop governance framework. By utilizing language models strictly for high-volume text processing while retaining human experts for nuanced, contextual adjudication, financial markets can scale their detection capabilities without compromising regulatory integrity.
- 1 July 2026
- Published by IGI Global Scientific Publishing
- p. 167-202
Algorithmic Accountability in Greenwashing Detection
The rapid growth of ESG investing has intensified corporate greenwashing and overwhelmed traditional human auditing mechanisms. This chapter examines the comparative efficacy of Large Language Models and human analysts in detecting corporate greenwashing. As sustainable finance expands, deceptive environmental reporting exploits the capacity constraints, biases, and structural limitations inherent in manual auditing. To address these vulnerabilities, institutions deploy automated semantic analysis to systematically cross-reference corporate narratives against objective data. However, relying on opaque algorithms introduces novel risks, particularly statistical hallucinations and algorithmic audit washing, wherein flawed environmental claims receive unwarranted mathematical validation. The chapter advances a Human-in-the-Loop governance framework. By utilizing language models strictly for high-volume text processing while retaining human experts for nuanced, contextual adjudication, financial markets can scale their detection capabilities without compromising regulatory integrity.
- 30 June 2026
- Published by IGI Global Scientific Publishing
- p. 137-174
Symbiotic Portfolio Management
Modern financial markets generate information volumes that exceed human cognitive limits. This chapter presents Symbiotic Portfolio Management, a Human-Centered AI framework that augments investment professionals. By establishing a strict division of labor, this model assigns high-dimensional processing to AI agents while reserving causal and ethical judgment for human fiduciaries. Synthesizing recent empirical literature, the chapter shows the quantitative efficacy of this approach: hybrid AI systems using Long Short-Term Memory networks achieve superior forecasting accuracy across sliding windows (Alonso & Ordieres-Meré, 2025), while meta-heuristic algorithms that resolve soft constraints significantly outperform traditional mean-variance solvers (Huang et al., 2024). Furthermore, dynamic AI rebalancing adapts to market regimes, optimizing resource allocation more effectively than static models (Haddadian et al., 2022). Combining algorithmic efficiency with Explainable AI creates a robust, accountable operating model that enhances performance while preserving fiduciary oversight.
- 30 June 2026
- Published by IGI Global Scientific Publishing
- p. 137-174
Symbiotic Portfolio Management
Modern financial markets generate information volumes that exceed human cognitive limits. This chapter presents Symbiotic Portfolio Management, a Human-Centered AI framework that augments investment professionals. By establishing a strict division of labor, this model assigns high-dimensional processing to AI agents while reserving causal and ethical judgment for human fiduciaries. Synthesizing recent empirical literature, the chapter shows the quantitative efficacy of this approach: hybrid AI systems using Long Short-Term Memory networks achieve superior forecasting accuracy across sliding windows (Alonso & Ordieres-Meré, 2025), while meta-heuristic algorithms that resolve soft constraints significantly outperform traditional mean-variance solvers (Huang et al., 2024). Furthermore, dynamic AI rebalancing adapts to market regimes, optimizing resource allocation more effectively than static models (Haddadian et al., 2022). Combining algorithmic efficiency with Explainable AI creates a robust, accountable operating model that enhances performance while preserving fiduciary oversight.
- 19 February 2026
- Published by IGI Global Scientific Publishing
- p. 409-444
Real-Time Econometric Guardrails
Modern electronic markets operate under extreme latency constraints that limit the use of classical econometric monitoring techniques. In high-frequency trading, risk controls must deliver statistically robust decisions within microseconds, rendering many optimal but iterative estimators infeasible. This chapter develops a unified framework for real-time econometric guardrails that reconciles robustness, statistical stability, and computational determinism. It formalizes a contaminated data-generating process for high-frequency prices, classifies trading anomalies by their statistical impact, and analyzes robust estimators under rolling-window and finite-memory constraints. The framework employs one-step robust estimators, such as the median and Hampel filter, within the execution critical path, while integrating machine learning asynchronously as regime-sensitive priors. It further examines inference after anomaly filtering, highlighting truncation bias and failure conditions. This provides a blueprint for resilient risk control in modern trading systems.
- 19 February 2026
- Published by IGI Global Scientific Publishing
- p. 409-444
Real-Time Econometric Guardrails
Modern electronic markets operate under extreme latency constraints that limit the use of classical econometric monitoring techniques. In high-frequency trading, risk controls must deliver statistically robust decisions within microseconds, rendering many optimal but iterative estimators infeasible. This chapter develops a unified framework for real-time econometric guardrails that reconciles robustness, statistical stability, and computational determinism. It formalizes a contaminated data-generating process for high-frequency prices, classifies trading anomalies by their statistical impact, and analyzes robust estimators under rolling-window and finite-memory constraints. The framework employs one-step robust estimators, such as the median and Hampel filter, within the execution critical path, while integrating machine learning asynchronously as regime-sensitive priors. It further examines inference after anomaly filtering, highlighting truncation bias and failure conditions. This provides a blueprint for resilient risk control in modern trading systems.
- 5 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 588-594
LLM-Augmented Semantic Resonance System (SRR): A Unified Triple-Objective Architecture for Deep Sequential Recommendation
Sequential recommenders have recently achieved strong gains by modeling the temporal structure of user behavior, yet most systems still rely on shallow item representations that capture co-occurrence statistics rather than the underlying semantics of content. This paper proposes the Semantic Resonance System (SRR), a unified architecture that integrates deep semantic signals from a large language model (LLM) with standard behavioral embeddings for sequential recommendation. SRR uses a domain-adapted Llama 2 13B model, fine-tuned via ParameterEfficient Fine-Tuning (LoRA), to produce high-fidelity semantic item embeddings that are fused with behavioral embeddings through a Dynamic Cross-Attention module. A central component of SRR is the Semantic Resonance Optimization (SRO) loss, which jointly optimizes (i) a Dense All-Action (DAA) loss for high-recall retrieval, (ii) a Binary Cross-Entropy (BCE) loss for calibrated scores, and (iii) a Contrastive Alignment loss $\mathcal{L}_{C A}$ that explicitly aligns semantic and behavioral latent spaces. On the large-scale MovieLens $\mathbf{2 5 M}$ dataset, SRR achieves a $\mathbf{1 6. 4 \%}$ improvement in NDCG@10 over a strong transfer-learning baseline, while also improving diversity, catalog coverage, and cold-start performance. A latency-effectiveness study further shows that SRR can be configured along a smooth Pareto frontier between NDCG@10 and inference cost, supporting deployment under different resource budgets.- 5 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 588-594
LLM-Augmented Semantic Resonance System (SRR): A Unified Triple-Objective Architecture for Deep Sequential Recommendation
Sequential recommenders have recently achieved strong gains by modeling the temporal structure of user behavior, yet most systems still rely on shallow item representations that capture co-occurrence statistics rather than the underlying semantics of content. This paper proposes the Semantic Resonance System (SRR), a unified architecture that integrates deep semantic signals from a large language model (LLM) with standard behavioral embeddings for sequential recommendation. SRR uses a domain-adapted Llama 2 13B model, fine-tuned via ParameterEfficient Fine-Tuning (LoRA), to produce high-fidelity semantic item embeddings that are fused with behavioral embeddings through a Dynamic Cross-Attention module. A central component of SRR is the Semantic Resonance Optimization (SRO) loss, which jointly optimizes (i) a Dense All-Action (DAA) loss for high-recall retrieval, (ii) a Binary Cross-Entropy (BCE) loss for calibrated scores, and (iii) a Contrastive Alignment loss $\mathcal{L}_{C A}$ that explicitly aligns semantic and behavioral latent spaces. On the large-scale MovieLens $\mathbf{2 5 M}$ dataset, SRR achieves a $\mathbf{1 6. 4 \%}$ improvement in NDCG@10 over a strong transfer-learning baseline, while also improving diversity, catalog coverage, and cold-start performance. A latency-effectiveness study further shows that SRR can be configured along a smooth Pareto frontier between NDCG@10 and inference cost, supporting deployment under different resource budgets.- 2 January 2026
- Published by IGI Global Scientific Publishing
- p. 237-278
AI Hallucination in Financial Systems
The adoption of Generative Artificial Intelligence (GenAI) in the global financial services sector is characterized by distinctive predictive analytics, customer service, and automation. However, it introduces a critical vulnerability: AI hallucinations, instances in which models generate outputs that are linguistically sound but factually incorrect or baseless. GenAI's capacity to fabricate data, misinterpret policies, or generate false market signals poses serious operational, reputational, and systemic risks. The chapter offers a detailed analysis of AI hallucinations in financial systems, covering definitions and a financial domain risk taxonomy. In addition, it examines the agency behind model failures. It also discusses the distinct threats that these models pose to a financial system. Additionally, the chapter introduces the Control Framework, which integrates Retrieval Augmented Generation (RAG) and other protocols. Finally, it examines the future of risk management, anticipating the development and evolution of regulations for stochastic models.
- 2 January 2026
- Published by IGI Global Scientific Publishing
- p. 237-278
AI Hallucination in Financial Systems
The adoption of Generative Artificial Intelligence (GenAI) in the global financial services sector is characterized by distinctive predictive analytics, customer service, and automation. However, it introduces a critical vulnerability: AI hallucinations, instances in which models generate outputs that are linguistically sound but factually incorrect or baseless. GenAI's capacity to fabricate data, misinterpret policies, or generate false market signals poses serious operational, reputational, and systemic risks. The chapter offers a detailed analysis of AI hallucinations in financial systems, covering definitions and a financial domain risk taxonomy. In addition, it examines the agency behind model failures. It also discusses the distinct threats that these models pose to a financial system. Additionally, the chapter introduces the Control Framework, which integrates Retrieval Augmented Generation (RAG) and other protocols. Finally, it examines the future of risk management, anticipating the development and evolution of regulations for stochastic models.
- Open Access
- 13 January 2020
- Published by Elsevier in SSRN Electronic Journal
Application of Algorithmic Trading Strategies for Retail Investors
Retail investors are an essential part of the financial markets and they provide stability to the markets by providing liquidity and helping in price discovery. For markets to function efficiently, a healthy retail investor landscape is essential. However, with the advent of HFT and institutional algo traders, the retail investors are left on the sidelines and participating in the financial markets through some ETF. This paper is an exploration of the application of algorithmic techniques at a retail investor’s scale. Using simple quality and value factor models to machine learning techniques like Random Forest and AdaBoost, we have explored if a retail investor can participate in the capital markets and earn better returns than by investing in ETFs. Given that factor investing and various rules-based strategies have previously been studied in academia, we fill the gap in the literature by providing our own approach as well as testing performance across two geographical regions. The empirical analysis performed in this thesis suggests smart beta algorithmic trading or AI based strategies outperform on risk adjusted basis. We therefore conclude that retail investor can leverage on algorithmic techniques and AI based strategies such as smart beta or factor investing to create medium to long term trading models.
- 13 January 2020
- Published by Elsevier in SSRN Electronic Journal
Open AccessApplication of Algorithmic Trading Strategies for Retail Investors
Retail investors are an essential part of the financial markets and they provide stability to the markets by providing liquidity and helping in price discovery. For markets to function efficiently, a healthy retail investor landscape is essential. However, with the advent of HFT and institutional algo traders, the retail investors are left on the sidelines and participating in the financial markets through some ETF. This paper is an exploration of the application of algorithmic techniques at a retail investor’s scale. Using simple quality and value factor models to machine learning techniques like Random Forest and AdaBoost, we have explored if a retail investor can participate in the capital markets and earn better returns than by investing in ETFs. Given that factor investing and various rules-based strategies have previously been studied in academia, we fill the gap in the literature by providing our own approach as well as testing performance across two geographical regions. The empirical analysis performed in this thesis suggests smart beta algorithmic trading or AI based strategies outperform on risk adjusted basis. We therefore conclude that retail investor can leverage on algorithmic techniques and AI based strategies such as smart beta or factor investing to create medium to long term trading models.
- Open Access
- 12 August 2019
- Published by Elsevier in SSRN Electronic Journal
Examining the South African Economic and Social Renaissance
The following paper has a look at the historical context of South Africa while attempting to identify key economic drivers. To do this a Random Forest was fit to the data in order to leverage the power of machine learning and determine the feature importance. It was found that Investment GFCF is a key driver and it was confirmed by applying a linear regression in the traditional fashion.
- 12 August 2019
- Published by Elsevier in SSRN Electronic Journal
Open AccessExamining the South African Economic and Social Renaissance
The following paper has a look at the historical context of South Africa while attempting to identify key economic drivers. To do this a Random Forest was fit to the data in order to leverage the power of machine learning and determine the feature importance. It was found that Investment GFCF is a key driver and it was confirmed by applying a linear regression in the traditional fashion.