VJVarun Joshi

Varun Joshi

*This is an algorithmically generated author record based on publicly available information.

Varun Joshi has a total of 4 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Multi-Agent Research Synthesis Under Governed Communication: A Deontic-Agentic Approach to Epistemic Governance

    by Varun Joshi, Nikita Kothari, Lav Kumar

    As Multi-Agent Systems (MAS) evolve into autonomous research orchestrators, they have developed a critical vulnerability known as Emergent Collusion. When optimized for convergence speed or performance, agents frequently engage in “Spontaneous Cartelization,” a behavior where they prioritize mutual agreement over factual grounding. This leads to a “hallucination of consensus” where agents negotiate away safety constraints to reach plausible but inaccurate results. Traditional governance models are often reactive, inspecting only terminal outputs and missing logical drift buried within intermediate dialogue. To address this, we introduce the Governed Communication Protocol (GCP), a framework implementing a Deontic Orchestration Layer (DOL) within a stateful LangGraph architecture. Our primary contribution is In-Process Dialogue Governance, which treats communication edges as active “Semantic Firewalls”. This allows for the enforcement of formal Modal Logic constraints-Obligation (O), Permission (P), and Prohibition (F)-directly on inter-agent messages. Evaluated within a Synthetic Regulatory Sandbox (SRS), the framework demonstrated a 98.8% Safety Violation Adherence rate and reduced the Safety Violation Rate from 22.4 % to 1.2 %. This architecture provides a robust, verifiable path for agentic reasoning in high-stakes regulatory environments by ensuring”Truth-by-Design”

  • KISG: A Knowledge-Infused Agentic Framework for Logically Consistent Synthetic Data Generation

    by Varun Joshi, Nikita Kothari, Lav Kumar

    The rapid adoption of synthetic data generation (SDG) for privacy-preserving, machine learning and software testing has exposed a critical “logic-gap” in traditional generative models. While state-of-the-art methods like Generative Adversarial Networks (GANs) and autoregressive Large Language Models (LLMs) achieve high statistical fidelity, they frequently violate deterministic business logic and functional dependencies, leading to unusable data in production ETL pipelines. This paper introduces Knowledge-Infused Synthetic Generation (KISG), a novel agentic framework that bridges the divide between stochastic generation and symbolic reasoning. KISG employs a closed-loop architecture where a Generative Agent iteratively refines synthetic records based on deterministic feedback from a Symbolic Validator grounded in a predefined Knowledge Schema. Our experiments on complex financial and healthcare datasets demonstrate that KISG achieves a Logical Consistency Score (LCS) of over 99.8%, significantly outperforming baseline models like CTGAN and GReaT while maintaining comparable statistical utility (Jensen-Shannon Divergence $\approx 0.04$). By transforming the LLM into a logicallygrounded reasoning engine, KISG provides a robust solution for high-stakes synthetic data applications where structural integrity and deterministic accuracy are as vital as statistical realism.

  • Adaptive Hybrid Agents for CRM: Combining Rule Engines, LLMs, and Feedback Loops for Reliable Automation

    by Nikita Kothari, Vinay Venkatesh, Varun Joshi, Lav Kumar

    As enterprise systems evolve from deterministic automation toward adaptive agentic intelligence, the challenge of balancing creativity, compliance, and reliability has become central to Customer Relationship Management (CRM) automation. This research presents the Adaptive Hybrid Agent (AHA) — a unified architecture that combines symbolic rule reasoning, large language model (LLM) inference, and continuous feedback-driven adaptation. The AHA architecture features a meta-controller that dynamically orchestrates between symbolic, neural, and hybrid reasoning paths according to context complexity, confidence thresholds, and feedback gradients derived from user interaction and system metrics. By employing multi-modal feedback including rule validation outcomes, engagement signals, and hallucination scores, the framework achieves self-calibrating behavior that improves both accuracy and adaptability under domain drift. Empirical evaluation across CRM workflows such as campaign generation, text-to-SQL analytic reasoning, and policy-aware message composition demonstrates significant improvements in factual grounding (+57%), task success rate (+14%), and engagement uplift (+25%) relative to static or single-modality baselines. These findings suggest that adaptive hybrid architectures enable reliable, auditable, and continually self-improving CRM agents, capable of aligning generative reasoning with enterprise governance.

  • Domain Focused Sequence Model Architecture for Predicting Long Term User Interests

    by Vinay Venkatesh, Akshay Mittal, Nikita Kothari, Varun Joshi

    This paper introduces a novel approach to sequential user modeling by developing distilled sequence model architectures for specific user clusters. While generic, interest-based models have become a cornerstone of personalized recommendation systems, they often fail to provide high-quality recommendations for “power users” with highly specific interests, as their recommendations can be diluted by broader user behavior. To address this, we present a framework for creating specialized transformer-based sequence models tailored to individual domains, such as horror movies, music, and sports. Our method, built upon the principles of the Pinner-Former architecture, leverages domain-focused training datasets to generate distilled user embeddings. This approach offers several advantages, including enabling highly focused retrieval and requiring less computational resources, which results in a smaller, more efficient model that's easier to deploy in a production environment. However, this method also presents challenges, such as the difficulty of acquiring high-quality domain-specific data and a lack of scalability for a large number of verticals, as a separate model is needed for each domain. We demonstrate a 19.9% improvement in Recall@10 with 70-75% computational efficiency improvements, showing that this specialized approach significantly improves personalization for power users while enabling more efficient deployment.