Green circular icon with a rightward arrow inside

Joshua Field

Solo founder building criminal defense AI on a knowledge actualization engine.

Three patent-pending AI products: GIDEON, [HUB], and UltrAI. CA Bar attorney. Bay Area.

GIDEON
Criminal defense AI

Built on the Atom-Structured Retrieval (ASR) system I created. ASR decomposes 241 California felonies into 126 atomic primitives with 100% formula uniqueness — the first codified standard for criminal defense reasoning.

GIDEON gets more intuitive and more strategic with time. It maintains context within a case, across a caseload, and factors in the workflows, pace, and tactics of the individual attorney using it. To best prepare a defender, it's not enough to know the law and the facts — you have to know the context. How to balance a crushing caseload. How to triage a crowded hearing calendar. How to spend your finite hours where they'll matter most for the client in front of you.

Privacy-first architecture: per-firm key derivation + cloud egress chokepoint. No individual case facts ever reach a shared model.

Provisional patent filed. Methodology paper public on GitHub.

241
California felonies decomposed
126
Atomic primitives
100%
Formula uniqueness
[HUB]
AI Co-Intelligence

A knowledge actualization engine. The partnership layer between you and AI, built on the Continuous Context Protocol (C3).

Connects to the tools you already use — Gmail, Calendar, Drive, Notion, code editors — and serves the right knowledge at the right time, in any token budget.

GIDEON is one vertical of [HUB], and the framework lends itself to more — each new build draws on the understanding and context that came before it, combining what's being built with what the engine already knows.

Provisional patent filed on the closed-universe baseline architecture.

1,841
Expertise distillates · 100% injection hit rate
9
Specialized departments
2,731
Indexed knowledge artifacts
UltrAI
Multi-LLM orchestration

Patent-pending system that routes prompts across multiple AI models (Claude, GPT, Gemini), runs multi-round refinement, and synthesizes outputs into a single high-quality response. Python and FastAPI. Live demo available.

About

I spent more than a decade as a public defender, campaign manager, and policy expert. As a defender alone: 1,200+ cases, sometimes 150 concurrent (three times the recommended max), 80%+ success rate at trial. The job teaches you one thing about the criminal justice system: defense has no codified standards. Prosecutors have a playbook. Defenders improvise. Running data-centric campaigns at scale taught me the other thing — that strong system architecture is what separates an operation that survives its own complexity from one that doesn't.

Before tech.Where I came from.

GovernmentObama Administration appointee (Peace Corps Press Director, secret clearance). Six campaign cycles.
OperationsPennsylvania Voter Protection 2016 — 3,000 volunteers, 2,000+ Election Day incidents, zero legal challenges. Fair Census Project — $2.5M budget, 0.06% variance.
AIMIT Certificate in Designing and Building AI Products and Services. Three production systems shipped. Provisional patents across the stack.
How I build
A particular method — proof-backed, fractal, partnership-first.
Proof-backed
Fractal
Partnership-first
Mechanical privacy
Evidence
Proof-backed

The architecture is grounded in a formal proof. 12 axioms, 5 theorems, 24 invariants, 19 failure modes accounted for. Most AI systems are "trust me." Mine is "check the proof." The proof does real work day-to-day: it keeps builds constrained to what the architecture actually permits, ties testing back to the user experience the system is supposed to deliver, and keeps maintenance targeted at the failure modes that actually matter.

Fractal

GIDEON (criminal defense) and [HUB] (knowledge actualization) aren't separate products — they're the same engine instantiated for different domains. Build the trunk right once; the branches inherit. Organizational and multi-firm scales are the same engine again at larger scope.

Partnership-first

I don't build AI to replace people. I build the partnership layer between people and AI. The system gets sharper the more you work with it — learning your patterns and surfacing the right information and the right next move in the form you'll actually use. Every interaction is captured, distilled, and brought forward, so the AI you're using a year from now is the one shaped by everything you've done together.

Mechanical privacy

Open-weight local models do the work that touches your data. Nothing leaves the device unless it passes through a single audited chokepoint. The patent on the closed-universe baseline architecture enforces this structurally — not policy, not promises.

Evidence

A particular method should produce particular results. Across nine patent experiments and ~960 controlled trials, the system architecture has been measured, not asserted.

  • Distillates beat raw transcripts on retrieval quality with large effect (Cliff's d > 0.7).
  • Friedman χ²(2) = 31.12, p < .001 across the three-way ranking comparison.
  • Context Diffusion outperforms BM25 baselines on multi-hop queries with statistical significance.
  • GIDEON ASR delivers 100% formula uniqueness across 126 primitives with 7.91-bit entropy.
  • All experiments pre-registered; multiple-comparison corrections applied (Bonferroni–Holm).
Get in touch