This AI Startup Ships Pipeline Changes in Days, Not Weeks | Kepler AI Case Study

Kepler's engineers were fixing pipeline problems about twice a week: stuck runs, schedules stopping, dead-lettered queues. Here's why they built their own orchestration system, and why they tore it out after four weeks with Prefect. Kepler builds enterprise AI for high-stakes industries, starting with investment firms, where every figure it returns has to trace back to a source document. Getting that right meant ingesting and processing millions of documents reliably, which meant the orchestration layer underneath had to just work. What we cover: - Why a homegrown orchestration framework was eating engineering time meant for the core product - How Kepler evaluated Dagster and Prefect, and what tipped the decision - The 4-week migration Where do things stand now? 170,000+ flow runs a week, 20M+ documents processed, orchestration maintenance near zero Try Prefect free: prefect.io Resources: Prefect docs: https://docs.prefect.io GitHub: https://github.com/PrefectHQ/prefect X: @PrefectIO If this was useful, subscribe!