Orchestrating Generative AI Workflows to Deliver Business Value
summary
Presents a practical framework for embedding LLMs into an existing Python stack using open source tooling: Metaflow as an orchestration and versioning layer with an @kubernetes decorator to send heavy jobs to cloud GPUs, a productionized parallel Stable Diffusion pipeline (Hugging Face, Docker, Nvidia GPUs, S3) and image→video work, Whisper deployed on Kubernetes, and retrieval‑augmented generation that pulls Metaflow docs to yield the correct decorator. Shows swapping Pinecone/LanceDB and Llama 2 replacement with almost no migration cost, and calls out fine‑tuning cost and quantization as concrete tradeoffs.
AI-written summary · about
chapters (7)
- 0:00Generative AI workflows intro
- 3:40Traditional vs ML stacks
- 11:25Metaflow for ML orchestration
- 17:27Generative AI risks
- 23:10LLM deployment challenges
- 28:53LLM best practices
- 31:43Q&A
related talks
27mAI-Powered ETL Pipeline Orchestration: Multi-Agent Systems in the Era of Generative AI
47mEmbedding AI: Don’t Burn Your Fingers in the Heat of the AI revolution
25mBuilding AI Workflows: From Local Experiments To Serving Users
34mThe State of AI and Machine Learning Operations in 2019
42mLLMs on Autopilot: Running AI Agents on Kubernetes With Open Source Tools
06mLightning Talk: Harnessing Generative AI Without Making a Faux Pas
30mContainer-enabled Asyncio is All You Need (to Build Pythonic AI Workflows at Scale)
13mLightning Talk: AI Meets Kubernetes: Langflow & GROQ in Automated Incident Management