NLP IL x Taboola - May 2025 Meetup

To get updates from Taboola click here.

Agenda:

18:00-18:30 - Gathering, food, and drinks
18:30-18:45 - NLP IL & Taboola : Opening words
18:45-19:15 - Keren Corsia (Taboola) -
Democratizing Data: How LLMs Make Complex Queries Easy for All
19:15-19:45 - Hai Rozencwajg (JLL) -
From Email Overload to Actionable Insights
19:45-20:15 - Orel Babayoff, PhD (Nimble) -
E2E Product Matching at Scale: Embedding Retrieval, GT Curation, and LLM Distillation

Abstracts:

## Lecture 1 - Democratizing Data: How LLMs Make Complex Queries Easy for All

Lecturer: Keren Corsia, Data Platform Software Team Lead
Lecture Abstract: In the age of Large Language Models (LLMs), the promise of democratized access to organizational data is more tangible than ever. Yet, in practice, many businesses struggle to turn this potential into reality. This talk presents Taboola's journey in building “Sage,” an internal query engine that enables business users to retrieve data using natural language—without needing to write SQL. One of the core challenges we faced was making sense of a highly complex data landscape, with over 7,000 tables in our primary data warehouse and multiple diverse data sources across the company. We explore the key challenges of integrating LLMs in enterprise settings: handling fragmented schemas, capturing business context, and ensuring query accuracy. Attendees will learn about the architecture of Sage, how NLP techniques like intent detection and context retrieval play a central role, and what it takes to bridge the gap between language models and real-world data. This session offers practical insights for NLP practitioners interested in operationalizing LLMs in complex data environments.

## Lecture 2 - From Email Overload to Actionable Insights

Lecturer: Hai Rozencwajg, Senior Data Science Manager at JLL
Lecture Abstract: Explore a cutting-edge application of LLMs revolutionizing industry news consumption. This presentation showcases a production-ready AI solution that transforms the daily flood of newsletters into a short feed of actionable insights.
Uncover the inner workings of an innovative system that processes thousands of news articles from diverse sources. Discover how LLMs are optimized to grasp complex industry contexts, pinpoint market-moving information, and forecast relevance with exceptional precision.
Key features include:
•⁠ ⁠A comprehensive prompt library operating seamlessly in the background
•⁠ ⁠Valuable insights on crafting optimal prompts for maximum effectiveness
•⁠ ⁠Sophisticated company matching using text embeddings
•⁠ ⁠Real-world impact: The transformative effect of AI-powered news filtering on decision-making processes
This session offers invaluable knowledge for data scientists, AI engineers, and industry professionals seeking to leverage AI for enhanced information management and strategic decision-making.

## Lecture 3 - E2E Product Matching at Scale: Embedding Retrieval, GT Curation, and LLM Distillation

Lecturer: Orel Babayoff, PhD, Head of AI at Nimble
Lecture Abstract: We share our journey in building a scalable, end-to-end product matching system—from ground truth curation and embedding-based candidate retrieval to LLM-assisted validation and model distillation. Alongside the matching logic, we’ll cover the custom data pipeline powering the process and how we balanced accuracy, speed, and cost to reach production scale.