Llama 2 & 3 Safety: Soumya Batra on Agentic AI Training

What if you could build AI features with confidence while moving at the pace of innovation? In this episode, Benjamin Wagner sits down with Rohit Girma, Staff Software Engineer at Airbnb, to explore how to evaluate generative AI in production, why breaking down complex problems into smaller chunks accelerates development, and the key strategies for scaling AI-powered products beyond zero-to-one. Whether you're shipping AI features or transforming your engineering workflow, this conversation offers practical insights on building reliable AI systems, leveraging LLMs as orchestration tools, and the future of software development. Tune in to discover why humans remain essential in the scaling phase and how your team can move faster without sacrificing quality.

In this episode of The Data Engineering Show, Benjamin sits down with Paul Tune, Staff Research Scientist at Canva, to explore the advancement of machine learning at one of the world's leading design platforms. Learn how Canva is transitioning from traditional ML like recommendation engines for templates to cutting-edge agentic workflows that allow users and AI to collaborate on complex design tasks. Whether you're interested in the infrastructure behind distributed training or the nuances of post-training LLMs for aesthetic tasks, this deep dive offers a masterclass in scaling ML for millions of creative users.

What if the expertise that built foundation models could reshape how you think about AI's future? In this episode, Benjamin sits down with Soumya Batra, founder and CEO of WisePort AI and former safety lead on Llama 2 and Llama 3 at Meta, to explore how foundation models evolved from traditional NLP, why post-training holds the highest leverage for safety and controllability, and what natively agentic AI means for the next frontier of AI development. Whether you're curious about the model training lifecycle or wondering what comes after large language models, this conversation unpacks the technical strategies and vision shaping tomorrow's AI systems.

What if building a distributed SQL engine meant rethinking everything about how query execution works at scale? In this episode, Benjamin sits down with Nikita, Senior Software Engineer at Cloudflare, to explore how R2 SQL leverages object storage and distributed computing to power analytics across 300 global locations, why backward compatibility becomes critical when you can't control infrastructure rollouts, and the key strategies for handling joins and adaptive query execution in a stateless, point-to-point network architecture. Whether you're designing distributed systems or curious about how Cloudflare processes petabytes of data, this conversation reveals the real-world engineering challenges and innovations shaping the future of cloud data platforms.

What if you could deploy ML features and real-time data pipelines without building complex infrastructure from scratch? In this episode, host Benjamin sits down with Nikhil Simha, CTO at Zipline AI and co-author of Chronon AI, to explore how Chronon, an open-source system that generates data infrastructure from simple queries, is transforming feature engineering at companies like OpenAI and Airbnb. Learn why iteration speed matters for fraud detection, how to serve thousands of signals at a massive scale, and what the future of analytical databases looks like in an AI-first world. Whether you're scaling real-time ML systems or building customer-facing analytics, this conversation is packed with practical insights on bridging the gap between data scientists and ML engineers.

What if your data platform could power both critical business decisions and real-time product features at scale? In this episode, host Benjamin sits down with Magnus Dahlbäck, Senior Director of Data and Platform at Voi, to explore how a metrics-first approach and semantic layers transform data accessibility, why traditional ML and LLMs require different strategies for different problems, and how to balance FinOps costs while processing billions of IoT events daily. Whether you're building data infrastructure for a high-growth company or rethinking how your organization consumes data, this conversation is packed with practical strategies for unlocking data value and preparing your platform for AI. Tune in to discover how Voi ditched traditional BI tools and revolutionized their approach to enterprise analytics.

What happens when a team of seven engineers spends a year trying to build a production-ready CDC connector and fails? For Artie CTO and co-founder Robin Tang, it was the spark needed to build a platform that makes data streaming accessible. In this episode, Robin joins Benjamin to discuss the "DFS" (Deep First Search) approach to data sources, the engineering hurdles of real-time Postgres-to-Snowflake pipelines, and why "theoretically correct" architectures often fail in practice.

What if your data platform could serve AI-native workloads while scaling reliably across your entire organization? In this episode, Benjamin sits down with Ritesh, Staff Engineer at Lyft, to explore how to build a unified data stack with Spark, Trino, and ClickHouse, why AI is reshaping infrastructure decisions, and the strategies powering one of the industry's most sophisticated data platforms. Whether you're architecting data systems at scale or integrating AI into your analytics workflow, this conversation delivers actionable insights into reliability, modernization, and the future of data engineering. Tune in to discover how Lyft is balancing open-source investments with cutting-edge AI capabilities to unlock better insights from data.

What does MLOps look like when you are deploying 60 billion machine learning predictions a day? Maddie Daianu, Head of Data and AI at Intuit Credit Karma, joins the Data Bros to pull back the curtain on one of the most high-volume data environments in FinTech. With a 100-person team serving 140 million members, standard data practices break down. Maddie shares how her team manages terabytes of daily data on Google Cloud and explains the massive strategic pivot they are undertaking right now: The move from "Information" to "Agency."

Nike’s Principal Data Engineer Ashok Singamaneni joins Benjamin and Eldad to discuss his open-source data quality framework, Spark Expectations. Ashok explains how the tool, which was inspired by Databricks DLT Expectations, shifts data quality checks to before the data is written to a final table. This proactive approach uses row-level, aggregation-level, and query data quality checks to fail jobs, drop bad records, or alert teams - ultimately saving huge costs on recompute and engineering effort in mission-critical data pipelines.