Research / Experiments

Questions before claims

I am interested in experiments that expose where a system works, where it fails, and what those boundaries cost in practice.

Reliable structured data extractionConditional acceptance

Deterministic pipelines vs. large language models

An experimental comparison of deterministic parsing and LLM-based extraction across clean and noisy inputs, with emphasis on extraction quality, latency, cost, and failure behavior. The work grew from the Vantage student-record ingestion pipeline.