Accelerating the Journey of an AI Algorithm to Production with OpenFaaS
summary
BT PLC cuts algorithm-to-production from years to 2 months, claiming 90% faster delivery by packaging models as web endpoints with OpenFaaS + Kubernetes. Pipeline uses simple YAML, builds in CI (Jenkins/GitLab), pushes containers to Harbor, and deploys in 12–15s; Prometheus and Elasticsearch collect metrics; reuse via a Function Store speeds iteration. Demo shows a code-quality pipeline delivered in three months, cold-starts around 2–2.5s, and on-premise or portable Docker deploys for trials.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:11BT Applied Research
- 2:11The Challenge...
- 3:35LifeCycle of an Algorithm
- 5:09Levels of Expertise
- 7:04Rapid Dev. and Reuse
- 10:41New Algorithm LifeCycle
- 13:34Cloud-Native Delivery
- 15:42Demonstration
- 23:26Serverless 1.0 Landscape
- 27:20Serverless Functions Made Simple
- 28:10Ecosystem demo
- 31:59OpenFaas Architecture
- 33:05BT Dashboards
- 33:44OpenFaaS (new) Watchdog
- 34:06Events
- 34:21Open Source feedback
- 35:31Impact to BT
- 36:53Future work
related talks
34mFaaS and Furious - 0 to Serverless in 60 Seconds, Anywhere
39mBuilding Serverless Application Pipelines
23mOpenFaaS: Serverless platform with no strings attached?
35mGetting Beyond FaaS: The PLONK Stack for Kubernetes Developers
19mDiscovering serverless with OpenFaaS
19mNo Token Budget: Agentic Software Factories on Kubernetes
28mOperating a Global-Scale FaaS on Top of Kubernetes
39mFaaS: A Contemporary, Comparative Study