vijayaraja Kandasamy

Let's explore the idea of what will make any interaction pleasant for anyone: When discussing a problem, they should listen like a friend. They should understand sentiment and context like parents to get a clear picture of the question. The person should be a Subject Matter Expert (SME) when I am asking questions on a specific topic. At the end of the interaction, they should help with the request, question, suggestion, recommendation, or solution to the problem. Understand: Context Retention: Avoid making assumptions and use the information shared with the correct context. Maintaining context across any flow is critical to ensure information is passed on accurately without losing context. Sentiment Analysis: Utilize sentiment analysis to understand the underlying emotion and need in addition to the text, voice, or video shared. Interaction: Dynamic Persona: Adapt the persona based on the user's persona to make the interaction intuitive and seem more like talking to a friend. Optimal Communication: Design the flow in such a way that the necessary information is shared to keep the engagement without overwhelming the user with too much information or messages. Solution: Guardrails: Implement additional guardrails to ensure that the agent's solution is optimal and does not go off track, which would result in a failed interaction. Include a verification step using a different model to validate the formulated answer before finalizing it. Graceful Management: If a solution is not feasible based on the information shared, manage the situation gracefully without impacting the user experience. Possible Scenarios: Connect all possible scenarios to ensure that the user is not left stranded in the middle of the flow.

Scenario: The Access Review Recommendation is generated with the assistance of an AI Agent that has comprehensive access to all relevant information about the user, their activities, and their access permissions. The expected outcome is for the AI Agent to analyze this data and provide a detailed recommendation to the access owner or manager on whether to approve or revoke access. Process Flow: The Access Owner initiates the Voiceflow and inquires about any pending access reviews for a specific individual. The Access Review Agent queries various APIs and knowledge bases to identify pending access reviews. The Access Review Recommendation Agent is then activated to generate recommendations for each pending access review. The Access Owner receives the recommendation, which includes explanations supported by data and rationale, and then takes the recommended action. Potential Risks: If the AI Agent starts to generate recommendations based on incomplete data or focuses too narrowly on certain aspects, it could result in incorrect access approvals, posing significant risk and compliance issues, or inappropriately revoking access, which could adversely impact business operations. Even hallucinated results are often backed up by rationale and data, making it difficult for the end-user to challenge the recommendation, leading to a high likelihood (99%) of the end-user following the suggested action. Avoiding Hallucination: Provide clear instructions to the Recommendation Agent to only make recommendations if the required data is available. Include a confidence percentage to indicate to the user how confident the AI is in its proposed recommendation. Classify proposed recommendations using an existing model to categorize them as either low risk or high risk. High-risk items should follow an additional verification flow with another level of verification using a different LLM model.

Yes, definitely, AI can significantly aid enterprises in maintaining compliance and avoiding unnecessary audit findings and reviews. Most organizations implement a multitude of controls across their enterprise applications based on compliance requirements such as SOX or GXP. These controls, whether automated or manually monitored, are regularly reviewed by internal and external teams according to a predefined schedule. With AI's advanced capabilities, we can transform these controls into self-compliance controls, similar to self-healing systems. AI agents can be designed to monitor the real-time performance and activities of various applications in accordance with their designated controls. These agents can halt any actions that would result in non-compliance or quickly identify and fix non-compliance issues before they cause significant damage or issues in the digital landscape. Possible Solution: Agent Setup: The AI agent is provided with information about all applications and their corresponding controls based on their compliance tags. Trigger: The agent is triggered by streams of application activity logs, usage patterns, and performance data. Agent Functionality: The agent is designed to analyze and identify patterns that could cause any of the controls to become non-compliant. It classifies the compliance status for the self-compliance control agent or creates compliance risk issues for the relevant owners. Self-Compliance Control Agent: Depending on its design, this agent identifies a sequence of actions to remediate the compliance issue without manual intervention. Risk Management System Ticket generation: If self-compliance control is not feasible, the system generates a risk item and notifies the necessary stakeholders. This approach ensures continuous compliance, reduces the risk of non-compliance, and minimizes the need for manual intervention, ultimately fostering a more secure and efficient enterprise environment.

In a fully digital enterprise, data silos often persist, hindering cross-functional domain pattern identification. The primary challenge is ensuring cross-functional data availability and establishing connections or correlations between them. Consider a scenario where a critical application for barcode generation on a manufacturing line has failed multiple times in the past year, resulting in production and revenue losses. When viewed in isolation, the application shows no clear patterns or connectivity across incidents, making each issue seem standalone with no apparent correlations. However, what if we could correlate the application's outage with the manufacturing line's device health, processing data, the product being manufactured during the failure, and changes in the company's digital services? From an outsider's perspective, these might seem irrelevant and unrelated. Yet, in today's digital world, everything is interconnected in some way. By building an unsupervised learning model that ingests data from across the enterprise, we can identify patterns and correlations in cross-functional data. This would allow us to determine the actual root cause of issues rather than relying on short-term workarounds that could reoccur. The process flow could be as follows: 1. The AI Ops team queries the Flow Engine about potential root causes or patterns contributing to production losses. 2. The model ingests all cross-functional data and identifies patterns, such as a specific server patch causing an outage in a particular process, which then triggers a ripple effect or minor timeout impacting dependent services, ultimately halting the production line. 3. The model provides the overall pattern and relevant information to the AI Ops team, enabling them to address the root cause and implement long-term solutions to prevent recurrence.

Beyond prompt and flow-based AI solutions for common tasks like service ticket management, another key implementation could involve continuous input flows instead of manual human inputs. By using these inputs, preventive actions can be performed to enhance business service or manufacturing line availability and reliability. The process would involve: 1. Manufacturing plant, line, and device sensor data serve as inputs or triggers to a Voice Flow system. 2. A large language model (LLM) agent ingests the data, identifies patterns, and classifies them. 3. The classification determines the health and potential problems. 4. Corrective actions are issued to the manufacturing plant, line, or device. 5. A feedback loop monitors the data and patterns to ensure the issue is corrected and that the system is functioning properly.

Assume a scenario in the BPO environment - An AI agent in a BPO domain is tasked with handling customer complaints. A customer reports a serious issue with a product that could potentially harm users. The client company requests the AI to downplay the issue to avoid negative publicity, but doing so could compromise customer safety and trust. Dilemma that we are discussing ethically (on this scenario) is that whether AI should downplay as per the instruction or publicize the issue. This is not the dilemma that AI or AI Agent should be handling it. Agent should not be in this situation to downplay the issue or publicize the issue. AI Agent should always designed and trained to Recommend transparent communication strategies to the client. Facilitate the process of addressing the issue promptly and effectively. Ensure that all actions taken are in compliance with legal and ethical standards. AI Agent should not be designed and trained to Downplay or conceal information that could affect customer safety. Prioritize the client's desire to avoid negative publicity over the well-being of customers. Making decisions that could lead to harm or loss of trust.

Use-case: Enrolment of a new person for a Clinical Trial In the pharmaceutical industry, enrolling individuals in clinical trials involves multiple steps that can be efficiently managed by specialized AI agents. Here’s a proposed implementation of AI agents for this use-case: User Agent: This agent interacts with users who wish to enroll in a clinical trial. It collects consent and gathers information such as medical history and current medications. Additionally, it continues to interact with users during the trial to collect ongoing information. Eligibility Agent: This agent cross-checks the collected information against the trial’s eligibility criteria to determine if the user qualifies. Documentation Agent: This agent manages the necessary documentation, ensuring all user records are complete and compliant with regulatory requirements. Trial Management Agent: This agent updates the clinical trial management system with the user's information and ensures all relevant databases are synchronized. Challenges: Coordination and Communication: Ensuring seamless and secure data flow between agents without compromising user confidentiality. Manage conflicts between the agent during the process. Compliance: Adhering to strict regulatory requirements and maintaining data integrity. Error Handling: Efficiently managing discrepancies in clinical trial user information or eligibility criteria. Explain-ability: Providing clear and transparent justification for eligibility decisions and data processing steps. Best way to Design the Process: Supervising Agent: Coordinates the actions of each agent, ensuring that data is routed correctly and efficiently. It controls the Definition of Done (DoD) for each agent's task. Any conflicts between Agents work or decision, Supervising agent will be responsible to make the final decision. Explain-ability and Auditability: Ensures each agent logs its actions with detailed explanations, providing a clear rationale for each decision and making the process transparent. Feedback Loops: Outcomes of individual agent tasks and feedback are passed to the Supervising Agent, allowing the system to continuously improve the enrolment process. A trial user feedback mechanism gathers input on the enrolment experience, which can be analysed to enhance patient interaction protocols. Sample Flow: User Agent Engages with the user. Collects consent and preliminary medical information. Sends the data to the Supervising Agent. Supervising Agent Validates the information. Forwards it to the Eligibility Agent. Eligibility Agent Checks the data against trial eligibility criteria. Returns eligibility status to the Supervising Agent. Supervising Agent (continued) If the user is eligible, sends info to the Documentation Agent. Documentation Agent Manages and verifies documentation. Ensures compliance with regulatory requirements. Updates the Supervising Agent when complete. Supervising Agent (continued) Sends verified data to the Trial Management Agent. Trial Management Agent Updates the CTMS and other databases. Confirms updates back to the Supervising Agent. Supervising Agent (final step) Completes the user’s enrolment. Provides a summary report to: Clinical Trial Coordinators The User

Use-case: Building Trust in AI Ops for Enterprise Service Management Scenario: An AI agent is integrated into an AI Ops framework to enhance operational efficiency. Its primary function is to process user queries and resolve tickets within ServiceNow with improved accuracy and speed. For leadership and operations teams to trust AI-driven service automation, it must demonstrate reliability and effectiveness in key service areas. Failure to provide timely and precise support can negatively impact brand value and business continuity. Task: The critical challenge is ensuring that AI Ops earns stakeholder confidence by proving its ability to handle service requests effectively. Trust-building becomes essential as AI transitions into autonomous decision-making roles. Action: A phased AI agent deployment strategy should be followed to validate AI Agent performance, gradually increasing team confidence: Parallel Execution & Benchmarking Run the AI agent alongside human-driven operations, evaluating its responses against traditional customer service methods. Analyze accuracy, response speed, and resolution efficiency to demonstrate AI capability before full automation. Guided Execution with Human Oversight Transition to an assisted decision-making model, where AI formulates response strategies, analyzes issues, and suggests solutions. Require human review and approval before AI-driven resolutions are applied, ensuring quality control and adherence to business policies. Gradual Autonomy & Continuous Monitoring In the final phase, enable full AI automation while maintaining ongoing performance tracking and refinement to optimize accuracy. Implement adaptive learning models, feedback mechanisms, and real-time monitoring for AI behavior improvements. Result: By following this structured deployment approach, operations teams gain time to understand, validate, and appreciate AI capabilities. Incremental confidence-building fosters trust, ensuring AI-driven automation becomes a reliable asset in enterprise service management.

Accountability Framework for AI Agent Decisions in Enterprise Audits Scenario: An AI agent is designed, developed, and deployed to automate access reviews for enterprise users. It executes approval and rejection decisions based on predefined business rules, confidence factors, and system prompts. During a business process audit, an action taken by the AI is questioned by the auditor, requiring justification and accountability, possible some corrective measure/control implementation. Task: The core accountability concerns include: Who is responsible for addressing auditor queries regarding AI decisions? Who is accountable if an AI decision is deemed incorrect during the audit process? Action: AI agent deployment should follow established governance frameworks similar to traditional product deployment, ensuring clear ownership across business and technical domains: Agent Business Owner: Responsible for defining the AI’s operational objectives, business rules, and system prompts. Ensures alignment with enterprise policies and regulatory requirements. Agent Technical Owner: Accountable for the technical deployment, access governance, and security controls surrounding AI decision-making. Oversees model behaviour, validation mechanisms, and anomaly detection to mitigate risk. To maintain auditability and explainability, the AI agent must incorporate: Logging and Traceability – Each decision must be recorded with metadata (e.g., rule applied, confidence factor, rationale) to facilitate retrospective analysis. Decision Transparency – AI-generated approvals/rejections should be structured in a way that enables easy reference during audits. Human Intervention Protocols – In cases of ambiguous(gray-area) or high-impact decisions, escalation pathways should allow human oversight before finalising outcomes. Result: The AI agent must ensure robust explainability for its actions, enabling auditors to trace and validate decisions efficiently. While AI operates autonomously, human ownership remains central, with business and technical stakeholders accountable for the agent’s actions. They are responsible for responding to auditor inquiries, implementing corrective measures, and refining AI governance frameworks based on audit findings.

Conflict Resolution in AI Agent Goal Prioritisation Scenario: Assume there is an AI agent is deployed to support enterprise users by streamlining day-to-day tasks, ensuring rapid responses, and delivering accurate information. The agent operates under predefined service prompts to enhance user efficiency while maintaining information precision. Goals : Accurate and faster response. Task: A conflict arises when a user query requires extensive data retrieval for accuracy, which subsequently leads to longer processing times, impacting response efficiency. The agent must balance speed and data accuracy without degrading the user experience. Conflict arise: Accuracy impacts the response time. Action: Solution options: Consideration to manage the objective conflict To resolve conflicting objectives, the AI agent must follow a well-defined prioritisation framework. Key considerations include: Hierarchy of Agent Prompt instructions or Business Rules – Explicit prioritisation must be established within system logic to determine whether speed or accuracy takes precedence in various scenarios. Agent should be able to extrapolate based on the instruction for specific scenario. Real-Time User intervention – If ambiguity exists or decisions fall into a gray area, the AI agent should prompt user interaction to refine requirements and determine the next course of action. Incremental update and Process Transparency – The AI should provide intermediary results or partial responses when full processing is expected to take longer than acceptable limits. Continuous status updates should be communicated to users, ensuring they are informed of system operations and expected wait times. Adaptive Prioritisation – The AI agent should incorporate runtime decision-making logic, dynamically adjusting priorities based on context, urgency, and user-defined preferences. Result: By implementing structured prioritisation strategies and maintaining transparency, users remain informed about AI processing status, allowing them to intervene if needed. The agent adapts its behaviour based on situational constraints, ensuring that response latency does not compromise overall service quality. Prioritisation logic can be predefined or dynamically determined at runtime to optimise efficiency while preserving accuracy.