2025-11-08 –, Talk Track 1
Most AI pipelines still treat models like Python UDFs, just another function bolted onto Spark, Pandas, or Ray. But models aren’t functions: they’re expensive, stateful, and unreliable. In this talk, we’ll explore why this mental model breaks at scale and share practical patterns for treating models as first-class citizens in your pipelines.
When data scientists move from prototyping to production, they discover that plugging a model into their pipeline is nothing like calling a UDF. Models bring unique challenges:
• Expensive (GPU hungry, rate-limited APIs).
• Stateful (versions, prompts, seeds).
• Unreliable (timeouts, OOM crashes).
Using real examples from multimodal pipelines, we’ll show why treating models as UDFs leads to brittle jobs, and introduce first-class patterns that solve these problems:
• One-line model loading in dataframes.
• Smart batching & caching without hand-tuning.
• Model-aware scheduling & retries.
• Row-level fault tolerance (99% completion instead of all-or-nothing failures).
Previous knowledge expected
R. Conner Howell is a software engineer at Eventual, Inc. where he works on Daft SQL. Prior to Eventual, he worked at Amazon on AWS CloudWatch Logs Insights and Amazon PartiQL – an open source SQL dialect for semi-structured data. His work interests include SQL, PL, and DBMS architectures. He is based in Seattle and enjoys cycling, climbing, and skiing the many mountains of the Pacific Northwest & BC.