N10 : Sentinel Core A Preemptive AI Breakdown Detection and Repair Model: Framework for Safer Smarter LLMs
| Students | Vivan Chowdhury
Nikit Rout |
| School | HDSB - W. H. Morden Public School - Oakville |
| Level | Junior 7/8 - Grade 8 |
| Group | Group 8 - Engineering and Computing II |
| Abstract | Major Large Language Models like ChatGPT frequently experience hallucinations and logical inconsistencies. While most mitigation strategies result in excessive refusal, Sentinel Core addresses these structural vulnerabilities through specialized JSON reinforcement grounding and precise refusal directives. Our experimental system aims to bridge the reliability gap, providing a safer and significantly smarter framework that ensures much higher dependability in accuracy for complex AI outputs. |