AI Memory Patterns: Save Tokens, Cut Costs | Chander Dhall | Azure Cosmos DB Conf 2026

AI systems must retain conversation history, tool outputs, and user context across multiple turns. Basic approaches can quickly inflate token usage or lose critical context. In this session, Chander Dhall (CEO of Cazton, 15-time Microsoft MVP) explores three memory patterns implemented using Azure Cosmos DB NoSQL: 1. Sliding Window Memory — summarization for recent and older turns 2. Hierarchical Memory — recent context, compressed history, and long-term tiers with intelligent retrieval 3. Entity-Based Memory Graphs — extracts and stores structured facts for precise recall You'll leave with concrete code patterns, guidance on selecting the right approach, and a reusable Cosmos DB schema for AI agent memory management. 👤 Connect with Chander Dhall 📝 Chander Dhall, CEO of Cazton, is a fifteen-time awarded Microsoft (AI) MVP, Microsoft Regional Director, Google Developer Expert, Azure Cosmos DB Cosmonaut and world-renowned technology leader in architecting and implementing solutions. In 2025, he was recognized as one of only twenty AI MVPs in the United States. He's not only rescued software development, cloud, big data, and AI teams, but also implemented successful projects under tight deadlines and difficult business constraints. His company, Cazton, has a proven track record of not just saving the client millions of dollars, but also providing expedited delivery time. Cazton's clients include Google, Microsoft, Thomson Reuters, Broadcom, AT&T, Dell, Bank of America, NBC Universal, American Express, Fandango, LinkedIn, VMware, McKesson, Macquarie Bank, and many other Fortune 500, mid-size and startup companies. In the field of AI, Chander has been at the forefront of building and deploying enterprise-grade solutions that leverage cutting-edge technologies like OpenAI's GPT models, Google's Gemini, Cohere's Command and Anthropic's Claude. He specializes in designing and fine-tuning generative AI models for real-world applications such as conversational AI, pred