About Liza
Liza Katz is a search, GenAI, and observability consultant with over 20 years of engineering experience. Whether leading ad-serving teams at ironSrc or improving performance, user experience and observability at Elastic, her work has always focused on turning large volumes of data into actionable insights. Today, at BigData Boutique, she helps companies optimize their search and research processes using GenAI. She’s also a mom to one curious little girl and three extremely opinionated cats.
Where to find Liza on the web:
- On LinkedIn: https://www.linkedin.com/in/lizak/
- On GitHub: https://github.com/lizozom
Let’s start from the beginning — how did you get involved in the search tech industry?
At Ironsource, I led teams working on one of the hardest problems in ad tech — choosing the best ad for each user in real time, at crazy scale. We were serving billions of ads per day, so performance had to be spot on. I spent a lot of time collecting telemetry, analyzing engagement, and building tools that turned all that data into insights — both for internal teams and for the algorithms making the decisions. After that, moving to Elastic felt natural. I joined as a tech lead, focused on making Kibana faster and more observable. But beyond the tech itself, what really excited me was helping customers solve real-world problems using search. Whether it was e-commerce, analytics, or logs — the goal was always the same: help people find what they need, quickly and reliably, no matter how much data they have.
Tell us about your current role and what you’re working on these days.
Today I work at BigData Boutique where I help customers build and deploy GenAI solutions that solve real problems. My focus is on improving agentic and semantic experiences — like smart chatbots, RAG pipelines, and hybrid search.
Could you describe a ‘favorite failure’—a setback that ultimately led to an important lesson or breakthrough in your work?
This story isn’t directly about search, but it taught me something I still apply every day.
Back at IronSource, I started noticing strange anomalies in our telemetry data. Some users would start watching an ad in one country and finish it in another. At first everyone — my team, my managers — assumed it was just people taking flights. But the cases kept growing, and it didn’t feel right. I spent weeks trying to reproduce it locally with no luck. I swapped libraries, rewrote ID generators, even implemented things from scratch — but the bug kept showing up only at scale.
Eventually, I stumbled across an obscure article describing a bug in Math.random() — a pseudo-random implementation that had been quietly copied around since 1997 (!). The fix was one line. But the lessons stuck:
When things get weird, throw away the layers. Strip out the abstractions until you really understand what’s going on.
Don’t assume that code you didn’t write — even from a “smart” source — is safe. Be a little suspicious. It’s healthy.
What are some of the biggest misconceptions about search that you often encounter?
One of the biggest misconceptions I see is people taking raw text, embedding it, and expecting everything to just work — as if embeddings are some magic fix-all. It rarely works that way.
Is there a log error/alert that terrifies/annoys you in particular?
Any log that doesn’t come with a stack trace 😅
If you're building something from scratch - what does your ideal search tech stack look like?
If I’m building something from scratch, I usually reach for Next.js. I like having both frontend and backend in one place, all in TypeScript — it keeps things clean and plays nicely together. I’m also glad server-side rendering made a comeback. After years of heavy SPAs slowing everything down, it’s refreshing to have fast, SEO-friendly pages again.
On the data side, I obviously have a strong preference for Elasticsearch and OpenSearch — but I don’t force it. I like combining different databases and letting each one shine where it fits best. Search engines are great, but they’re not always the whole story.
What is the most unexpected or unconventional way you’ve seen search technologies applied?
One unconventional use of search tech I saw during my time at Elastic was Kibana using Elasticsearch as a kind of relational database — storing things like dashboards, visualizations, and settings. We all knew it wasn’t ideal, and there were plenty of discussions about switching to something more suitable. But since it worked “well enough,” it just stayed that way.
Give us a spicy take/controversial opinion on something related to Search
Here’s a spicy one: most search problems aren’t technical — they’re domain problems. People rush to tune relevance scores or swap vector models, but skip the most important part: understanding what users actually need. Without that, you’re just optimizing noise.
Also — semantic search is great, but it’s not magic. Embeddings won’t save you if your chunks are garbage or your use case is fuzzy. You still need to think. You still need intuition. And yes, you still need to understand your data like a human being.
