DevOpsDays Cairo 2025 :Abdulrahman Abouzaid Senior Software Engineer AEMS ai, Egypt
summary
Shows that production apps break when AI adds similarity search at scale: Pinterest example with ~1.5 billion images and ~1M daily uploads causes SQL lookups to time out. Demonstrates a Python prototype that stores text as 768-dimension vectors in Pinecone vector DB, builds an HNSW index and returns nearest neighbors with similarity scores (Dark Knight example moves from 65% to 73%). Recommends shard by similarity, add data locality and autoscaling to cut latency and make retrieval practical.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:29Traditional databases limitations
- 3:41What is similarity search
- 5:06Understanding vectors
- 7:55Vector databases solutions
- 9:27Practical demo setup
- 12:37Demo problem demonstration
- 15:54Demo results and analysis
- 17:20Vector indexing explained
- 19:30Architecture revolution
- 21:15Data locality and scaling
- 22:22Indexing as the secret sauce
- 24:31Summary and Q&A
related talks
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
50mBuilding AI ready applications
11mSimilarity Search Application utilizing Vector types on vCore-based Azure Cosmos DB for MongoDB
13mBuilding High-Performance AI Literature Processing
12mBuilding Generative AI Apps with Azure Cosmos DB
50mHow to design and build AI applications with vector search using Azure OpenAI & Azure Cosmos DB
06mMachines and the magic of fast learning
26mEngineering Excellence: Platforms and Architectures for Real-Time AI Analytics