Memory and Retrieval Architecture in Long-Running AI Agents
Persistent memory architecture separates production AI agents from demos that fail in the field.
Marcus Oyelaran
Senior Contributor
Marcus Oyelaran
A former backend engineer turned independent researcher, Marcus has spent over a decade stress-testing search APIs across fintech and media platforms. His writing focuses on the practical limits, failure modes, and design tradeoffs engineers encounter when building production-grade search layers.
4 stories
Web Source Attribution and Citation in LLM Agent Outputs
Source attribution in AI agents must be built into retrieval pipelines, not added after generation.
Marcus Oyelaran
Chunking Strategies for Long-Form Web Documents in RAG
Multiple chunking strategies work best depending on document structure, not one method for all.
Marcus Oyelaran
Web Grounding to Reduce LLM Hallucination in Agentic Tasks
Tying agent outputs to retrievable sources stops hallucinations from cascading.
Marcus Oyelaran