追踪、评估、上线：AI Agent 生产实践

Anyone can demo an AI agent. Running one in production — where it's traced, evaluated, and trusted at scale — is a different discipline entirely. Join **ClickHouse and Women in Tech SG**, hosted at the **PayPal office in Singapore**, for an evening built for the engineers doing e

时间: 10月6日 周二 · 18:00

地点: PayPal Singapore

Anyone can demo an AI agent. Running one in production — where it's traced, evaluated, and trusted at scale — is a different discipline entirely. Join **ClickHouse and Women in Tech SG**, hosted at the **PayPal office in Singapore**, for an evening built for the engineers doing exactly that. We'll dig into the unglamorous work that makes agentic systems shippable: tracing every step an agent takes, building evals that catch regressions before your users do, and the real-time data infrastructure that keeps observability fast when traces hit the billions. **Who should come:** This one's pitched at a senior audience — AI/ML engineers, data and platform engineers, and technical leads shipping (or scaling) LLM-powered systems in production. If you're past the tutorial stage and deep in the "why did my agent do *that*" stage, this evening is for you. 📍 **Venue:** PayPal Office, Singapore 🗓️ **Date:** Tuesday, October 6, 2026 🕕 **Doors open:** 6:00 p.m. SGT **🗓️ AGENDA:** * :00 PM: Registration, Food & Chitchat * 6:30 PM: Welcome and Introductions by Amrita Mishra, Country Director, Women in Tech Singapore * 6:40 PM: **Talk - Trace. Eval. Ship: Running AI Agents in production** by Yisam Lee, Senior Support Engineer @ ClickHouse * 7:15 PM: Panel Discussion on "Engineers shipping payments at scale" * 7:45 PM: Wrap-up and Networking **🎤 Session Details: Trace. Eval. Ship: Running AI Agents in production** Every AI agent in production generates a flood of questions: which prompt version caused that regression? Why did the agent pick the wrong tool? Is the new model actually better, or just different? Yisam tackles these head-on, showing how end-to-end tracing turns agent behavior from a black box into queryable data, how eval pipelines built on real production traces replace gut feel with hard signal, and why the storage layer matters more than most teams expect — with a look at how platforms like Langfuse lean on ClickHouse to keep tracing and evals fast at billions

打开 cancan App 找同去的人