Data Challenges for Scaling Multi-Agent GenAI Apps with Python
summary
Practical takeaways and concrete trade-offs: use LangGraph or Semantic Kernel but expect them to mis-handle mixed operational+vector workloads; embed vectors into the same JSON record via change feed + Azure Functions embedding and observe the vector written back to the document; handle consistency with session token helpers when multiple agents read/write; expect disk ANN and sharded indexes to hit billions of vectors at sub-100ms if partitioning is right; consider store vectors and transactions together only after testing partition key and failover behavior.
AI-written summary · about
chapters (10)
- 0:00AI apps and GenAI intro
- 3:32Why LLMs matter
- 7:45Retrieval augmented generation explained
- 10:12Vectors and transactions storage
- 12:42Scaling in Azure Cosmos DB
- 16:37CAP theorem choices
- 23:39Vector and hybrid search in Cosmos DB
- 28:30AI agents and multi-agent systems
- 32:44LangGraph multi-agent demo
- 34:09Q&A
related talks
32mApproach to Architecting Generative AI Solutions
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
48mHow to Make Smart Architecture Decisions when Building Gen AI Apps
20mOwn your agent
24mScalable GenAI in Production
46mAI Agents for the Rest of Us
22mGenerative AI application stack and providing long term memory to LLMs
46mDesigning Semantic Memory for Multi-Agent Systems with Python