Back to Home
Research.
My academic publications, pre-prints, and explorations into the mathematics and architecture of emerging AI models.
Pre-PrintSeptember 2026
Evaluating Context Retention in Long-Context Large Language Models for Enterprise Data
Sumit Shrivastava, Co-Author Name
Abstract
"This paper explores the degradation of information retrieval in Large Language Models when context windows exceed 100k tokens. We propose a hybrid chunking-and-reranking mechanism that improves exact-match retrieval by 18% in proprietary corporate datasets..."