QGI Launches the Deterministic AI Stack: Five Products for Regulated Decisions

Today we are relaunching QGI as a product company. We are not a research lab that happens to publish software. We are a software company with a specific thesis — that AI used in regulated decisions needs a different architecture than AI used in consumer products — and a specific stack that delivers on it.

The thesis is simple. In a classical "RAG + LLM" pipeline, two calls with the same prompt can produce two different rationales. In a consumer chat, that is an interesting quirk. In a denied mortgage, a rejected insurance claim, or a blocked KYC case, that is a liability. It is a CFPB letter, a state Attorney General inquiry, an SIU deposition. Probabilistic AI does not survive audit. Deterministic AI is what the regulated future of AI looks like — and the stack has to be deterministic by construction, not by bolt-on guardrail.

Five products. One reasoning graph.

The QGI stack has five layers, plus Enterprise Blueprints as the delivery surface:

1. L1 Q-Prime

   Foundation · Encoding

   Quantum embeddings. The first commercial quantum-structured embedding model. Encodes enterprise data into a hypergraph that preserves polarity, scope, and cross-rule dependencies. Runs on NVIDIA CUDA-Q — no QPU required.

2. L2 QAG Engine

   Platform · Reasoning

   Chunkless RAG for regulated decisions. QAG reasons over whole documents instead of chunked retrieval. A Hilbert-Space Compacting layer projects document states into seven interpretable signals before generation, so contradictions surface as explicit audit signals.

3. L3 Qualtron

   Platform · Generation

   4M-context composite model. Specialized small models that compose into a 4M-token working context for regulated generation. Replaces general-purpose LLMs inside QAG where domain precision beats raw scale.

4. L4 Quantum Graph Memory

   Platform · Memory

   Time-aware graph memory. Graph memory for QAG agents. Preserves provenance and temporal structure of every fact and revision, so reasoning stays consistent across sessions and audits.

5. L5 Neural Symbolic Agents

   Runtime · Orchestration

   Custom agents for FS workflows. The runtime layer of the stack. Orchestrates underwriting, claims, fraud, and compliance with persistent memory, dependency tracking, and conflict coordination that survives audit replay.

The delivery surface is Enterprise Blueprints — mostly forks of NVIDIA AI Blueprints, wired to Q-Prime and the QAG Engine through the QGI Enterprise Factory. Documents QAG — the fork of NVIDIA's RAG reference blueprint, remade deterministic — is live today.

Why this positioning, why now

The company has been building this stack for more than a year. What changes today is how we describe it. We are retiring the "research lab" frame. We are retiring "mission-critical intelligence" as the marquee phrase. The new positioning is specific: deterministic AI for regulated decisions. That is the category we are building. It is the category our customers ask for. It is the category that survives audit.

Three things drove the timing. First, active enterprise evaluations in mortgage compliance — including evaluations disclosed in our March 2026 news releases — gave us concrete proof points inside regulated AI Governance frameworks. Second, Q-Prime reached HuggingFace and the QAG Engine opened public preview in April, so the stack is no longer a roadmap document. Third, the regulator landscape has hardened. State AGs, the CFPB, and prudential regulators are now asking for explainability at the individual-decision level — not at the aggregate-model level. Probabilistic AI is not going to clear that bar. Deterministic AI has to.

The three pillars of the relaunch

Availability — what is live, what is shipping, what is next

Industries and engagement

Financial Services is the anchor vertical, with active enterprise evaluations in mortgage compliance and credit workflows. Insurance is on the 2026 roadmap with an open waitlist. Named evaluations disclosed in news releases carry the verbatim third-party statements.

Neural Symbolic Agents — the top layer of the stack — ships as a custom enterprise engagement, not a self-serve product. We run a small number of pilots per quarter. Each follows a structured path: Scoping → Pilot → Audit → Production, with governance artifacts, model cards, signal-level dashboards, and replay tooling from day one.

Trust posture

Determinism is one half of the contract. Trust posture is the other. Our Trust page lays out the full picture: deployment models (Managed, VPC, On-premise), data handling, the compliance roadmap (SOC 2 Type II in progress, HIPAA-ready architecture, FedRAMP on the horizon), and the principle the stack is built against — if a decision matters, it must be understood.

What changes for existing evaluations and pilots

Nothing changes for active evaluations. Enterprise evaluations disclosed in news releases continue on their engagement path. Pilots already in scoping continue on their engagement path. The 5-product stack was the architecture being built behind the scenes; today it becomes the public interface. If you are already running a QGI pilot — or evaluating one — your contact, your workflow, and your contract stay the same. You now have named products to point at in your governance review.

About QGI

Quantum General Intelligence Inc. (QGI) is a San Diego-based AI company that builds the deterministic AI stack for regulated industries. QGI's five products — Q-Prime, the QAG Engine, Qualtron, Quantum Graph Memory (QGM), and Neural Symbolic Agents — are delivered through Enterprise Blueprints (mostly forks of NVIDIA AI Blueprints wired to Q-Prime and the QAG Engine). Co-founded by Dain Ehring (CEO, previously founder of Dorado) and Dr. Sam Sammane (Chief Scientist, PhD in Formal Verification). Learn more at qgi.dev or contact press@qgi.dev.