Lightning Talk: We Built a Unified AI Asset Catalog in Backstage
summary
Built a Backstage-based AI asset catalog that models inference servers as components and individual models as resources, hosted on an OpenShift cluster (GPU and non‑GPU nodes) and exposed through an API Gateway per server. TechDocs store access/login and usage instructions; OpenAPI specs import automatically and model cards are copied from Hugging Face. Pain points: duplicate model entities when modeling models as components, catalog RBAC is all‑or‑nothing, and templates cannot dynamically discover models. Extensions to auto-ingest models and custom views are underway.
AI-written summary · about
chapters (8)
- 0:00Unified AI asset catalog intro
- 1:52Modeling models as components
- 3:18Initial dependency challenges
- 4:22Inference servers as components
- 6:28What worked well
- 7:51Roadblocks and limitations
- 8:55Next steps and extensions
- 9:57Conclusion
related talks
21mLeveraging GenAI for Seamless Developer Experiences in Backstage
23mAgentic Backstage: How To Manage an AI Software Catalog
09mBackstage Beyond Engineering Teams: Can We Deliver Value to Other...
58mWorkshop: Empowering Collaboration: AI Developer Experience - Your Bridge from Mod...
42mDefense against the dark arts of AI: The playbook for a sovereign model-as-a-service platform
48mThe Room of Requirements for Developers: Scaling AI Magic with Internal Portals
24mWhat We Learned During Our Journey in Using Backstage at DAZN
17mChallenges and takeaways of managing AI workloads on cloud