Mahesh Goud

• Performed Human-in-the-Loop (HITL) validation of AI-generated document classification outputs across a 100K+ document enterprise workflow — reviewing model decisions for accuracy, consistency, and governance alignment. • Reviewed AI model outputs for misclassification patterns and explainability gaps; documented findings and escalated control gaps to stakeholders with structured recommendations. • Aligned AI-assisted processes with Responsible AI principles and NIST AI RMF and ISO/IEC 42001 frameworks in day-to-day validation and governance workflows. • Evaluated tools and systems supporting AI and data workflows for governance and compliance risks, contributing to control gap identification and remediation planning. • Developed and maintained Power BI dashboards to track AI output QA metrics, accuracy trends, and risk indicators — supporting data-driven governance oversight and reporting. • Contributed to governance documentation and audit trail improvements, supporting examination readiness under emerging AI governance requirements.

Responsibilities: • Supported operational risk processes, compliance checks, and documentation reviews • Ensured BAU accuracy through periodic audits and analysis • Performed data cleanup and classification support activities • Strengthened control processes and improved documentation standards • Supported cross-functional coordination for risk and compliance initiatives Achievements: • Consistent top performance with high-quality scores • Improved process accuracy through gap analysis and remediation

Verified identity, SSN, address & loan data for mortgage & servicing portfolios. Ensured accuracy of underwriting & compliance documentation. Supported initial borrower risk review. Achievement: Reduced documentation errors through improved verification checks.

Reviewed identity documents, loan agreements, appraisal reports & closing disclosures. Ensured correct borrower–loan–property mapping for audit readiness. Supported Tesla vehicle-loan documentation reviews. Achievement: Delivered consistent top-quality scores across multiple quarters.

• Performed adversarial prompt testing on LLM models (GPT-2 / GARAK framework) — designing and executing prompt injection and jailbreak scenarios to evaluate model safety boundaries and output behaviour. • Analysed model responses for hallucination risks, unsafe outputs, and risk amplification across model responses; produced structured findings aligned with responsible AI and AI safety assessment principles. • Developed a structured approach to assessing model risks and vulnerabilities, informing practical understanding of governance control requirements for LLM systems.