Onboarding new sellers onto retail platforms like Walmart and Amazon involves a complex, multi-step process designed to mitigate fraud and ensure compliance with global regulations. One of the most critical and cumbersome steps is Know Your Customer (KYC) verification, requiring sellers to upload documentation for identity verification, business registration, and compliance checks. This manual review process often leads to long approval times and delays, frustrating legitimate sellers and creating operational bottlenecks for compliance teams.

To address these challenges, we leveraged foundational models with custom prompting strategies, in-document summarization, and retrieval-augmented generation (RAG) to ground responses in trusted data sources, powered by open-source LLM APIs. By automating document analysis and augmenting human reviewers with AI outputs, we reduced overall onboarding time by more than 20 percent, improving seller experience and operational efficiency.

However, deploying AI into a regulated process like KYC required a robust responsible AI framework combining scalability with governance. We implemented guardrail models to flag edge cases and ensure human oversight, enforced strict data anonymization protocols to protect sensitive information, and applied privacy-preserving techniques for model training. We also established a rigorous validation pipeline to test outputs against regulatory standards, mitigating risks such as hallucinations and interpretability gaps.

This talk offers actionable insights for data scientists, compliance officers, regulators, and machine learning practitioners working at the intersection of AI, risk management, and regulatory compliance. Attendees will walk away with a practical framework for deploying AI in sensitive domains covering risk management strategies, scalable AI architectures aligned with compliance, and key lessons on balancing innovation with accountability.