QGI Logo QGI
Public preview Platform · Reasoning

The QAG Engine

Quantum-Augmented Generation platform

The reasoning layer of the QGI deterministic stack. Contradictions surface as one of seven named signals before any model generates a rationale. This is what replaces retrieval-centric RAG in regulated decisions.

The seven HSC signals

Hilbert-Space Compacting projects reasoning into seven interpretable scalars

Each signal is a property you can log, threshold, alert on, and review in isolation. A compliance reviewer does not need a data-science team to interpret a Conflict = high, Coverage = low state — the semantics are first-class, not buried in a latent space.

1 Signal

Relevance

Is this rule or data on-topic for the decision at hand? The scalar most systems already report — but here it is the first, not the only, signal.

2 Signal

Conflict

Where do two rules, overlays, or sources disagree? Contradictions surface explicitly so compliance can review them — not hidden inside a latent score.

3 Signal

Overlap

Which rules cover the same situation with different language? Overlap signals pre-empt the 'duplicate rationale' failure mode in multi-source retrieval.

4 Signal

Redundancy

How much of the retrieval is re-stating information already encoded? Redundancy signals keep the reasoning trace tight and auditable.

5 Signal

Coverage

Is every rule, overlay, or condition relevant to the decision represented at all? Coverage flags gaps before generation — no 'silently missing rule' failures.

6 Signal

Coherence

Do the retrieved fragments form a self-consistent narrative? Low coherence warns that the source corpus itself may have gaps or inconsistencies worth flagging.

7 Signal

Topology

What is the shape of the reasoning graph? A wide, shallow topology reads differently from a deep, narrow one — and a regulator can see which was used.

Composition

Read together

The signals compose into a decision-ready gate. The QAG Engine proceeds to generation only when the signal profile matches policy — or it raises the explicit state back to the workflow for human review.

Capabilities

What the engine is

Core

Hilbert-Space Compacting (HSC)

The core innovation. HSC projects high-dimensional quantum-structured states into seven interpretable scalars the engine consumes before generation. Reasoning becomes inspectable at the signal level — not at the model level.

Property

Deterministic reasoning contract

The same input produces the same reasoning trace every time. No temperature knob that quietly turns your compliance decision into a different answer tomorrow. The determinism is a contract the engine enforces, not an output-tier knob.

Audit

Replayable inference

Every decision is stored with its Q-Prime encoding, HSC signal values, and reasoning path. When a regulator or auditor asks to replay the decision, you hand them the trace — not a reconstruction.

Pricing

Commercial tiers

Startup, Growth, Enterprise, and OEM commercial tiers. Preview access is free during the public preview window. Contact us for pricing and tier selection.

Positioning

Supersedes classical RAG

Where a RAG pipeline asks an LLM to compose over retrieved fragments, the QAG Engine asks the LLM to generate over a reasoned structure. Contradictions are resolved before the model speaks — which is where most RAG failures in regulated workflows live: upstream of generation, in silent retrieval gaps and unresolved contradictions.

Compatibility

Works with any LLM

QAG is a reasoning platform, not a model. You can ship it with your existing LLM provider or pair it with Qualtron when Qualtron ships. The signals you get out are the same.

Pairing

Integrates with Q-Prime

The QAG Engine reasons over Q-Prime's quantum-structured encoding. Using another embedding model works, but you lose the structural guarantees that make the seven HSC signals meaningful.

Access

Public preview available

The public preview opens access to evaluation-grade deployment. Production-grade pilots run as scoped engagements in regulated workflows — claims, credit, compliance, AML/KYC. Request access below.

Enterprise proof path

Built for enterprise AI Governance reviews

QAG Engine pilots are scoped around a real business decision, the policy corpus behind it, and the audit trail a risk team must defend. Evaluation language stays precise until a customer, partner, or deployment is formally announced.

Request QAG Engine preview access →
QAG Engine FAQ

What is QAG and how is it different from RAG?

QAG stands for Quantum-Augmented Generation. Classical RAG retrieves fragments and asks an LLM to compose over them — most failures in regulated workflows are upstream in that retrieval step, where silent gaps and contradictions never surface. QAG replaces the retrieval step with structural reasoning over Q-Prime's quantum hypergraph, then projects the result into seven interpretable signals (the HSC layer) before any generation happens. Contradictions surface as signals, not as hallucinations.

What exactly are the seven HSC signals?

Relevance, Conflict, Overlap, Redundancy, Coverage, Coherence, and Topology. Each is a scalar that can be logged, thresholded, and reviewed independently. A 'Conflict = high, Coverage = low' state, for example, is a reviewable condition before your model ever generates a rationale.

How do I get preview access?

Use the Request Preview Access button above. Tell us which workflow you want to evaluate and we will follow up with the preview terms. We onboard a small number of evaluation partners per month.

Does the QAG Engine work without Q-Prime?

It can operate over other embeddings, but the seven HSC signals rely on the structure Q-Prime preserves. Without Q-Prime, you get deterministic orchestration but lose the interpretability layer that makes conflict and coverage reviewable. Most customers deploy Q-Prime and the QAG Engine together.

Move past probabilistic RAG.

Request preview access and we will help you set up an evaluation in a scoped regulated workflow.

Request preview access →
Partner with QGI