Projects

Research, not a portfolio.

Each project here answers four questions: what problem it addresses, how it approaches that problem, what was actually built, and what evidence exists that the work holds up.

Active · submitted for external evaluation

The Epistemic Stack

An adversarial pipeline that turns a contested debate into a typed, verifiable dependency graph instead of a confident-sounding paragraph, so the structure of a disagreement can be checked by a third party, not just read.

Read the case study
Research · pilot audit

Epistemic Calibration Audit

A pilot audit of how social pressure, authority cues, and contextual framing affect the judgments produced by frontier language models.

View project
Open research toolkit

Protocol Drift Atlas

An open toolkit for measuring protocol sensitivity and identifying changes in language-model safety judgments across evaluation conditions.

View project
Open research · audit framework

Aletheia Plurality Audit

A research framework for auditing plurality in AI-generated deliberation, examining how model outputs represent agreement, disagreement, and competing perspectives.

View project