Data Engineering Podcast

Keerti Kulkarni, an Amazon technical program leader, reveals how telemetry and AI automate physical logistics across more than 100 facilities. Explore petabyte-scale event streams, privacy and reliability challenges, AWS data pipelines, conversational troubleshooting agents, predictive maintenance, cost attribution, and safer AI rollouts in complex warehouse environments.

Christian Bremeau, Meta Integration Technology founder and metadata standards veteran, joins CTO Simon Dynin, a specialist in lineage and semantic modeling. They explore why fragmented tools create metadata silos, how deep lineage and versioning support trustworthy analytics, and why AI agents need governed semantics. Topics include MetaKarta’s vendor-neutral Semantic Hub, metadata operations, change impact, compliance, and the future of semantic compilation.

Christopher Doidge, founder of UNI Consulting and author of Where Are the Insights?, explores Agile Ledger Architecture, a practical approach to reducing data debt. He discusses making gold-layer data genuinely useful, exposing hidden logic in analyst scripts, preserving business definitions in durable ledgers, and applying the 15-minute test for time to insight. He also examines AI-ready data, incremental adoption, and when ALA is unnecessary.

Soham Azumdar, co-founder and CEO of Wisdom AI with experience at Google and Rubrik, explores what “context” means for modern data and AI systems. They discuss semantic layers, tribal knowledge, query logs, agent memory, and connecting data across silos. The conversation also covers validating, pruning, measuring, and keeping context fresh so AI agents can deliver trustworthy insights.

Yetunde Dada, Senior Director of Product Management at Astronomer who leads work on Airflow and Otto. She talks about Otto, an AI agent adding Airflow-specific context for DAG authoring, failure investigation, upgrades, migrations and scheduler modernization. The conversation covers model choices, security and integration with existing tooling, plus a roadmap toward human-approved self-healing pipelines.

Ragnor Comerford, engineer and entrepreneur building OmniGraph, a lakehouse-native graph layer for agentic systems. He explains graphs as semantic world models and why combining graph semantics with Git-style branching enables shared context, safer collaboration, and governance. The conversation covers architecture choices, agent-first query design, tradeoffs versus analytic engines, and emerging uses like company knowledge systems and agent orchestration.

Prukalpa Sankar, co-founder of Atlan and builder of metadata and context layers, explains building a “context flywheel” for AI agents. She discusses why model smarts need institutional, semantic, procedural, and tool knowledge. Conversations cover bootstrapping context from warehouses and BI, evolving catalogs into context layers, simulations and governance loops, and agents becoming a company brain.

Jevin Maltais, a fractional CTO and staff-level engineer who built product-driven streaming at Zapier, Humi, and Clio. He talks about practical Kafka use and common misuses. He covers treating events as source of truth, KTables and materialized views, schema registries and type safety, CDC with Debezium, and making Streams/Connect accessible through config-as-code with clear escape hatches.

Shravan Gunda, founder and CEO of Kaarvi, builds an AI-native, agent-driven platform to automate ingestion, quality, transformations, and dashboards. He explains Kaarvi’s multi-LLM parallel approach for reliability. Hear about synthetic data for rapid testing, text-to-SQL and text-to-dashboard chat workflows, domain-specialized agents, on-prem vs SaaS tradeoffs, and a roadmap for a marketplace and desktop assistant.

Weimo Liu, co-founder of PuppyGraph and former engineer at TigerGraph and Google F1, builds zero-copy graph querying for lakehouses. He discusses an edge-sharded, vectorized MPP design that avoids ETL. The conversation covers mapping tables to graphs, Iceberg-aware caching, integrations with MongoDB, and real-world uses like cybersecurity and entity resolution.