Showing cs.AIShow all
3 papers · 1 filter
cs.AI2026
Do Synthetic Personas Predict Real Audience Response? A Headline-Ranking Study with a Generic-Reader Baseline
Alexandre Cristovão Maiorano
Do demographic personas improve an LLM's prediction of audience engagement? We compare a fixed ten-persona panel with a generic-reader baseline on Upworthy headlines, with suppleme…
cs.AI2026
Isolating LLM Alignment from Regex: Zero Coverage and Metric-Dependent Divergence Under Adversarial Mutation
Alexandre Cristovão Maiorano
Production LLM applications commonly stack a regex filter in front of model-side alignment; prior work found no measurable coverage gain from adding a live Gemini backend behind an…
cs.AI2026
LLM Readiness Harness: Evaluation, Observability, and CI Gates for LLM/RAG Applications
Alexandre Cristovão Maiorano
We present a readiness harness for LLM and RAG applications that turns evaluation into a deployment decision workflow. The system combines automated benchmarks, OpenTelemetry obser…