Data Agents
Beyond the Demo
Monday, November 23, 17:30 - 20:00
Wix Campus TLV
//---About our meetups
Data agents can look impressive in a demo. The real engineering challenge starts when they need to work reliably across complex systems, diverse data, and production environments.
In this meetup, we’ll explore how Wix engineers are taking data agents beyond the demo from operating across 8,000+ Airflow DAGs and resolving production incidents, to understanding domain-specific context and catching data issues that traditional pipeline monitoring can miss.
agenda
17:30-18:00
Gathering: Pizza & drinks
18:00-18:30
850 Engineering Hours Back Every Month: Wix’s Context-Aware Airflow Agent
Learn how Wix’s Data Group built a context-aware agentic platform to generate compliant code, debug local pipelines, and autonomously resolve production incidents across 8,000+ Airflow DAGs.
Grounded in proprietary infrastructure context, governance rules, and automated incident response workflows, AirBot triages failures, traces lineage, and opens automated PRs, saving 850 engineering hours every month.
Yarden Wolf
Data Engineer, Wix
18:30-18:40
Break
18:40-19:10
The Last 80%: Trust, Context and Friction in Production Data Agents
Building a data analysis agent that delivers an answer is the easy part. The real challenge is making it reliable across dozens of domains, each with its own data, terminology, KPIs and domain experts.
In this talk, we'll explore how we built DORA to understand diverse contexts, evaluate its confidence, surface assumptions, and know when to say, "I'm not sure." We'll also look at how domain experts help the agent improve, and how every interaction builds context that makes its answers better over time.
You'll leave with a practical blueprint for building trustworthy data agents at scale, and a clear understanding of why getting the answer is just the first 20% of the work.
Jonathan Ohana Bikel
Data Engineer, Wix
19:40-20:00
Networking
Green Is Not Done: Why You Should Add an Agent In the End of Your Pipeline
Most of our tooling watches pipelines that fail. But some of the costliest surprises come from pipelines that succeed: the job is green and the numbers are wrong, or the numbers are right and nobody noticed they moved.
In this talk, we'll look at what happens after the last task turns green, and why we started to put an agent there. We'll cover how it finds the slices you missed, how it tells a data issue from a business change, and how it learns from its own past verdicts, not only from the people who use it.
You'll leave with a blueprint you can add to the end of your own pipelines.
Eden Bar-Tov
Co-Head of the Data Engineering Guild, Wix
19:10-19:40
19:45-20:30
How Do You Know Your AI Agent Works? Building EvalForge 👷♀️
As AI shifts from a supporting tool to the core of our products, the key question changes - from "does it work?" to "how do we know it’s working well?"
In this talk, by Or Goldreich, we’ll introduce EvalForge - a platform for systematically evaluating AI agents across test scenarios, capabilities (skills, MCPs, sub-agents, rules), and assertions - from LLM judges to tool invocation checks - turning agent behavior into something measurable, debuggable, and comparable.
Beyond the technical architecture, we’ll explore why AI evaluation is fundamentally different from traditional software testing, the hidden challenges teams encounter when agents operate in production, and what changed once we could actually measure reliability, decision-making, and failure patterns at scale.
Or Goldreich
Full-Stack Developer, Wix
//---Speakers
meet our
speakers
Yarden Wolf
Data Engineer, Wix
Jonathan Ohana Bikel
Data Engineer, Wix
Eden Bar Tov
Co-Head of the Data Engineering Guild, Wix
//---Register
RSVP
Yunitsman St 5, Tel Aviv - Yafo
Monday, November 23, 17:30 - 20:00
For detailed navigation instructions check out this guide
