Case Study: Kumo AI accelerates model training failure diagnosis with Deductive AI

Kumo cuts training workflow triage from hours to minutes with Deductive AI

Kumo AI, a platform for building predictive models from relational data, faced significant challenges in diagnosing failures within its complex, Temporal.io-orchestrated model training workflows. Engineers spent hours manually tracing nested activities and correlating telemetry data to find the root cause of failures, which slowed resolution and impacted customer reliability. This ongoing challenge motivated Kumo to partner with Deductive AI to accelerate diagnosis and recovery.

Deductive AI implemented an AI-powered on-call agent that automatically traces workflow graphs and correlates data from systems like Temporal, Grafana, and Loki to pinpoint the root cause of failures. The solution delivers actionable summaries directly to Slack, drastically reducing triage time. Early results showed promising improvements in how quickly engineers can understand and respond to failures, turning reactive debugging into proactive reliability engineering.

Kumo