Personal Project · Nov 2026
Triagent
I increased incident-severity accuracy from 38% to 93% and correct abstention from 0/7 to 7/7 on 32 labeled incidents by developing a hybrid-retrieval RAG pipeline and LLM agent with code-enforced citations. I enhanced a 1.5B model’s severity accuracy from 28% to 88.5% on 200 incidents through
