Speaker John Ash

Building a system where truth has more power than money

SF Bay Area · speakerjohnash@gmail.com · speakerjohnash.com

Profile

I am an independent engineer, researcher, and writer working on the infrastructure for truth-seeking and collective sensemaking. My research practice, called Cognicism, has spent twelve years on a single question: how can a society learn, over time, who has been worth listening to, and decide together on that basis. The work I take on now is where the mission and the math both matter: civic infrastructure, public sensemaking, regenerative finance, and the institutions trying to keep open inquiry alive in an algorithmic age.

Areas of Focus

• AI for truth-seeking and the compression of signal from noise
• Source reliability and trust modeling over time
• Explainable AI, FourThought, and provenance-aware reasoning
• Collective intelligence and structured public sensemaking
• Regenerative economics and incentive design
• Applied LLM systems, agentic workflows, and tool use
• Time-series modeling and learned temporal encodings

Selected Engagements

By John Ash · Sole Proprietorship · 2017 to present

The vehicle for everything I do independently. It carries the research (Cognicism, the Iris framework, Ŧrust, the FourThought ledger, the Cognicist literature), the engineering practice (AI systems design and consulting for civic, ecological, and cultural clients), the public speaking and performance work (keynotes, panels, podcasts, workshops, live shows), and the creative output sold through byjohnash.etsy.com: oracle decks, prints, original art, apparel, music, and software.

The Society Library · Engineering Contract · 2024 to present

The autonomous research infrastructure that lets a small team build and maintain Society Library’s debate maps at scale. Work covers the full internal toolset and the public-facing surfaces it feeds:

• Junto, the chat interface with live debate map editing that researchers use to grow maps from claims, quotes, and citations.
• Fact checking pipeline and the agents that find verbatim supporting and refuting evidence for any claim under examination.
• Bifocal browser for web research with live-generated semantic indices.
• Claimgen, the deepest map-generation system in the Society Library suite (alongside the Reddit maps agent and Junto chat). Given a question, it searches the web and constructs a structured model of the collective debate around it: positions, counter-positions, arguments, and the citations behind each.
• Muse of Truth, an ElizaOS plugin that watches conversation threads, identifies claims that need checking, runs them through the Junto fact-checking API, and replies with a verdict: supported, refuted, partially supported, or rejected.
• Open Junto (github.com/SocietyLibrary/open-junto), the open-source release of three research tools and the shared streaming agent loop that powers them. Lead engineer on the release. See Open Source and Research Projects below.
• Society Library research tools, the broader internal toolset built on the Open Junto foundation, used by Society Library staff and collaborators for evidence gathering, claim work, and debate map construction.

Open Civics · Research Contract · 2026

Small ongoing research engagement in the civic technology ecosystem.

Regen Network · AI Consultant · 2020 to 2021

Built Ceres, applying the Iris framework I had been developing since 2017 to Regen’s regenerative economy. Distilled Regen’s core concepts, eco-credits, carbon markets, inter-subjective verification, ecological state machines, into a queryable language model interface that could answer stakeholder questions, advise on land stewardship, and generate original regenerative narratives. Ceres demonstrated, before fine-tuning was common practice, that a language model could carry the wisdom of a mission-aligned organization.

Open Source and Research Projects

Rhizome

A cognicist research agent base. The shared skeleton for agents that learn over time in text space. FourThought-native, with three execution patterns for one-pass tools, chat synthesis, and chat over a mutating artifact. Provider abstraction across Anthropic, OpenAI, and local Ollama models, prompt caching, audit logs, cost tracking, blueprint-style Flask integration, and a SQLite-backed local ledger with a roadmap toward federated learning across instances. Every other project in the stack, finance, health, literature, civic research, sits on top of it.

Open Junto

A research platform for building agentic research tools. Pluggable module architecture, a shared BaseAgent loop with WebSocket streaming, a shared tool registry that modules compose from, cost tracking, secure API key management, and a SPA shell that auto-generates a module dashboard. The Society Library research tools are built on this same foundation. github.com/SocietyLibrary/open-junto

Iris Studio

A creative production studio built on Rhizome. Image and post generation, semantic search over a corpus of philosophical art, analytics over engagement, and an agentic research surface that can plan, generate, and evaluate work across the publication pipeline.

Profit

A personal finance and capital allocation platform built on Rhizome. Connects to live accounts through Plaid and runs specialized agents over the resulting ledger: a bookkeeper that proposes entity rewrites on incoming transactions, a runway agent for cash flow planning, Lyra for chat, and a research agent that generates timestamped predictions with confidence and reasoning. Predictions are scored against price outcomes and fed back through Brier and direction accuracy.

Self Health

Personal health agent built on Rhizome. Conversational intake, longitudinal logging, and reasoning over the user’s own timeline.

Mind

The original Prophet codebase, 2017. TruthModel is a variational generative text model with a Gaussian latent space, trained to generate new thoughts by sampling from the prior. Months before “Attention Is All You Need.” The config directory includes a truth-scoring spec with speaker embeddings, sinusoidal time embeddings, truth, dissonance, and a scoring rule that rewards correct contrarianism. Ŧrust, written down eight years before the proof.

Core Skills

Machine Learning. Applied ML, probabilistic modeling, classification, time-series, learned temporal encodings, ablation methodology.

Deep Learning. PyTorch, TensorFlow, custom training pipelines, transformer architectures, fine-tuning, embeddings.

LLM Systems. Agentic workflows, tool use, prompt caching, provider abstraction, RAG, evaluation.

Backend. Python, Flask, SQLite, DigitalOcean, streaming, websockets.

Research and Design. Architecture for collective intelligence and reasoning systems, value modeling, narrative generation, deliberative design.

Education

University of California, Davis. Major in Psychology, minor in Communication. Graduated with high honors.