Staff Software Engineer and Platform Architect with 20+ years of experience designing intelligent, cloud-native, distributed, and compliance-critical platforms across LinkedIn, Credit Suisse/UBS, Fidelity Investments, Publicis Sapient, IBM, and public-sector institutions. Architect of LinkedIn Finance's first compliance-certified Azure platform and high-throughput automation systems supporting financial operations. Combines software architecture, AI-enabled workflow engineering, distributed systems, cloud security, API/platform design, and technical leadership, with hands-on expertise in Java/Spring, Python, TypeScript, React/Next.js/Angular, Kafka, GraphQL/gRPC/REST, Azure, AWS, and AI-assisted automation.

Title: Evaluating Agentic AI: Scorecards, Shadow Mode, and Promotion Gates - Abstract: Building an autonomous AI agent is easy, knowing when it is safe to ship is not. Offline benchmarks tell you little about how an agent behaves when it can call real tools, spend real budgets, and take irreversible actions. This talk presents a practical framework for deciding when an agentic system is ready for production. We introduce three complementary mechanisms: scorecards that measure capability, safety, oversight, and cost on a common scale; shadow mode, in which the agent runs against live traffic without acting, so failures surface as evidence rather than incidents; and promotion gates, explicit, measurable thresholds an agent must clear to advance from evaluation to production. We trace a single agent through all three, showing the artifacts each stage produces and the failures each one catches. Attendees will leave with a concrete checklist for gating autonomous agents into production, grounded in metrics rather than intuition.