Harshit Agarwal

Title of the Talk: From Rules to Agents: Autonomous AI for Real-Time Monetization Decisions at Marketplace Scale

Abstract:
Marketplace revenue systems — governing fees, promotional discounts, and seller incentives — rely on deterministic rule engines that satisfy correctness, auditability, and regulatory compliance but optimize decisions in isolation. As machine learning augments these rules, a coordination problem emerges: multiple models fire simultaneously on the same seller, draw from the same promotional budget, and produce individually optimal but collectively incoherent outcomes. No individual model can resolve this. It requires a coordination architecture.
This talk introduces an agentic AI framework in which three specialized agents share one promotional budget pool and each answer a distinct causal question. FeeAgent is always-on: it proposes a personalized base selling fee rate per seller, augmenting the platform’s static fee rules with a learned, seller-specific rate calibrated to long-term retention. PromotionAgent is periodic and conditional: it runs campaign-style “do X, get Y” offers — list more items, try a new category, adopt a feature — targeting specific short-term behavioural changes. RewardAgent gives direct financial value with no condition required — wallet credits, cashback, growth program enrollment — triggered at milestones or retention signals. Each agent submits bounded proposals to a shared coordination layer that enforces budget limits, regulatory fee caps, and fairness requirements as hard constraints, and produces a two-layer audit trace without exposing model internals.
The talk covers agent anatomy, the fee-engine-versus-agent boundary, a four-phase migration path from static rules to full multi-agent deployment, and four open research problems — including agent robustness under adversarial seller strategy, which connects directly to the author’s paper at this conference. Context is drawn from eBay’s global marketplace, operating across 14 million active sellers.

Bio

Harshit Agarwal is a Staff Engineer with 14+ years of experience designing and scaling distributed systems across global commerce, cloud, and telecommunications. At eBay, he serves as a technical lead for the Plutus monetization platform, supporting 112 million transactions daily and 3.18 million sellers. His expertise spans backend architecture, pricing and promotions, AI-accelerated engineering, and large-scale system integration, with multiple innovation awards and proven business impact.