Three projects in AI analysis, forecasting, and personal context.

I work on systems where AI personas debate structured questions, predictions are validated against real outcomes, and personal data stays under user control.

Projects

Perspectives

Multi-agent forecasting and decision analysis.

Decisions under uncertainty benefit from structured analysis that surfaces competing considerations. Perspectives runs eight AI personas through a multi-phase debate: blind proposals, interrogation, discussion, and ranked-choice voting. The output is a single PDF report.

Predictions are tracked against real outcomes and prediction market data. This feedback loop identifies which analytical approaches are well-calibrated and adjusts their influence in future analyses.

getperspectives.app

Universal Automation Wiki

World model infrastructure.

Predictions and analysis benefit from structured representations of how systems work. The Universal Automation Wiki models processes at every scale, from individual daily routines to company operations to government policy.

Open source, intended as shared infrastructure.

universalautomation.wiki

Dossier

Privacy-first personal context for AI interactions.

AI interactions improve with personal context (goals, preferences, risk tolerance, topic boundaries), but that context should stay under user control. Dossier stores everything locally. Inferences are generated during sessions and discarded unless the user confirms them.

Open source, to make the architecture verifiable.

Coming Soon
How they connect

The three projects address different layers of context for AI-assisted analysis. The Universal Automation Wiki provides structured representations of how systems work. Perspectives analyses what might happen within those systems through multi-agent debate. Dossier provides the personal context that determines what matters to the person asking.

The long-term direction is an ecosystem where predictions draw on structured world models, are stress-tested through forced debate, and are contextualised to the individual. Each project is useful independently. The connections between them make the analysis more grounded.

Three layers of context
Universal Automation Wiki
How systems work. Processes and structures at every scale.
Perspectives
What might happen. Analysis and prediction through structured debate.
Dossier
Who is asking. Personal context, owned and controlled by the user.
Design decisions
Structured disagreement

The debate architecture forces AI personas to challenge each other's reasoning through blind proposals and structured interrogation. Analysis from systems designed to disagree tends to surface considerations that a single model would not.

Validated against reality

Predictions are tracked against outcomes and prediction market data. This feedback identifies which analytical approaches are well-calibrated and which consistently over- or under-estimate.

User-controlled data

Personal context makes AI more useful. The trade-off should be explicit. Dossier uses local-first architecture with consent-driven sharing and ephemeral inference by default.

Open and transparent

Personal identity systems and world model infrastructure are open source to allow verification and collaboration. The decision-making engine is proprietary.

Writing

Recent posts