Booking.com - Tel Aviv - Machine Learning & AI Meetup

🎤 Featured Sessions
Teaching Models to Generalize: Data-Centric Recipes for Domain-Adaptive Travel LLMs
Speakers: Alon Berliner & Amit Meitlin, Machine Learning Scientists @ Booking.com
Organizations deploying LLMs in production face a persistent tension: third-party APIs offer generality but come with latency, cost, and data governance concerns, while fine-tuned in-house models deliver superior performance on company-specific tasks, along with speed, cost efficiency, and control, yet tend to be brittle, degrading when input distributions shift or label spaces evolve. In this talk, we present techniques developed at Booking.com to close this generalization gap, producing fine-tuned models that remain robust under real-world change. We evaluate generalization along two axes that matter most in production settings: the ability to correctly handle unseen labels introduced after training, and the ability to maintain performance on out-of-distribution inputs originating from new domains or product surfaces. Our results show that these techniques, applicable across classification and structured-prediction tasks with minimal adaptation, allow in-house models to achieve the flexibility traditionally reserved for general-purpose APIs, without sacrificing their inherent advantages in performance, latency, cost, and privacy.
Scaling Agentic AI at Booking.com: From Real-Time Agents to Offline Pipelines
Speakers: Ofri Kleinfeld & Ilan Khirin, Machine Learning Engineers @ Booking.com
Building a single AI agent is easy. Running thousands of agent configurations across millions of records -reliably, observably, and with consistent quality is a different problem entirely. We built an agentic platform and SDKs that serve 40+ GenAI use cases across both real-time experiences and massive offline enrichment. The real-time layer provides a code-first SDK for authoring agents with configurable strategies (function-calling, ReAct), composable multi-agent workflows, tool catalogs, and memory patterns — powering trip planning, smart search, and proactive traveler support. The offline layer takes those same agent configurations and executes them at scale, processing millions of properties and conversations with self-serve backfills, streaming and batch output delivery, and per-use-case observability.
Optimizing Accuracy, Cost and Latency in Real-World Agents
Speaker: Or Dagan, Chief Product & Strategy Officer @ AI21 Labs
Most agentic systems rely on hardcoded heuristics to navigate execution decisions (e.g. which models, tools, and test-time compute scaling approaches to use) leading to efficiency leakage across cost, latency and accuracy. AI21 Maestro optimizes agents by learning to predict success, cost and latency probabilities across diverse actions and contexts, and driving runtime orchestration that intelligently navigates the full agentic action space. We will demonstrate how this approach yields state-of-the-art results and a new Pareto frontier on challenging agentic benchmarks, as well as the process required to optimize production agents.

🚀 RSVP
Space is limited – RSVP now to secure your spot!
We look forward to seeing you there!
📍 How to find us:
Booking.com Office, PWC Tower, Menachem Begin 146, 30th Floor, Tel Aviv Yafo
Enter the building from the middle entrance; someone will be there to direct you.
Recommended parking lot if you're arriving by car: Azrieli Town, Menachem Begin 146, Tel Aviv
Waze link: https://waze.com/ul/hsv8wrz8rk