Using Kaskada to Do Real Time AI
Talkat Open Source Summit Europe 2023 · on Wednesday, 2023-09-20· at 11:25· Room 0D-2-0D-3 (Floor 0), Abandoibarra Etorb., 4, 48011 Bilbao, Bizkaia, Spain·
summary
Kaskada is an open source, Rust/Arrow-powered realtime feature engine that claims native time travel for ML, lets you do timeline joins and windowed aggregations in Python, and benchmarks up to an order-of-magnitude speedup versus a fast baseline. The demo creates a RAG pipeline in about five minutes: Astra (Cassandra) stores 1,500-dim Ada-002 embeddings, Python queries nearest neighbors and calls OpenAI to answer (example returns “1920”). A small Slack bot project (beepGDP) and vector search in Cassandra prove the pattern works end to end.
AI-written summary · about
chapters (12)
- 0:00Kaskada and real-time AI
- 4:54What is Kaskada
- 6:43Time-series data with Kaskada
- 11:42Predictive ML with Kaskada
- 12:43Generative AI and RAG
- 14:53Live RAG chatbot demo
- 23:10Real-time + generative AI
- 26:18Slack bot example
- 30:30AI democratization challenges
- 31:12Changing app stacks
- 35:51Modern data platforms
- 39:45Cassandra 5 features
related talks
41mUsing Kaskada to Do Real Time AI
33mUsing Kaskada to do Real Time Machine Learning
40mOvercoming data infrastructure limitations with a new paradigm for ML
19mBuild Real time, AI powered Apps
18mKeynote: Data = AI: Leading the Future of Gen AI with Apache Cassandra
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
47mReal-Time AI with Open Source
26mOne Pipeline to Rule Them All: Unifying Multimodal and AI Data Processing with Daft