Reproducible AI with Kubeflow, lakeFS and Langchain
Talkat Cloud Native + Kubernetes AI Day NA 2024 · on Tuesday, 2024-11-12· at 15:20· Salt Palace | Level 1 | Grand Ballroom A·
summary
Builds a RAG chatbot from Project Gutenberg texts stored on S3, vectorizes chunks into a local vector DB, and queries OpenAI for answers while demonstrating a real failure where a scraped menu makes Wendy serve a Baconator. Uses Kubeflow + lakeFS + LangChain to replace an S3 loader with a ref-based data loader, commit and branch data in lakeFS, run CI checks to block bad merges, and keep canonical IDs so one can trace a bad response back to the exact data commit.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:45Reproducibility in AI
- 2:49LLMs logic and data
- 5:21RAG pipeline with LangChain
- 8:51Live demo
- 9:55Fixing data errors
- 12:00LakeFS for data versioning
- 13:49Updating pipeline with LakeFS
- 15:36Reproducible data pipelines
- 18:38Testing data quality
- 19:21Randomness in LLMs
- 20:05Versioning AI stack
- 20:46AI reproducibility best practices
- 21:08Q&A
related talks
01h 12mBuild Your Own AI Chatbot with OpenAI & LangChain
25mLangChain with Azure AI: Building LLM powered Applications
51mEvent Driven AI: Supercharging ChatGPT with RAG & LangStream
40m4 pillars for building a modern AI application in JavaScript
38mCraft your own Generative AI Chatbot with Amazon Bedrock
16mBuild a RAG app in minutes using Langflow OpenAI and Azure
01h 59mLLM applications with LangChain
01h 13mBuilding Production-Ready AI Systems in 90 Minutes