Quality Report Intelligence
Product documentation
Evidence-grounded quality reports from approved records.
From approved quality records. Every released factual claim cites a retrievable source. Preflight is a rule engine, not a model. The package returns to the QMS you already run.
Interactive demo: workflow UI you can run today. PoC backend and customer integration are engagement-scoped. Roadmap items are labeled in docs.
In the demo
Claim ledger, source locks, deterministic preflight, role-separated approval, QMS return failure modes, eight report templates (demo pack versions).
Roadmap (not current)
HWPX/DOCX/PDF-A export, OCR and multi-format ingestion, redlines and version compare, submission tracking, reporting analytics, customer-specific template packs.
What QRI is
QRI writes the quality documents your customer demands (NCR/CAR, 8D, FRACAS, government QA defect reports, counterfeit-part reports, QA activity plans, FAI/AS9102, PPAP) from evidence you already hold, with every released factual claim traceable to its source.
It is not a QMS. Your existing QMS stays the system of record. QRI imports the case, drafts and proves the report, gates its release, and returns the package.
Three commitments define the product. Grounding is enforced, not promised: generated text is a ledger of atomic claims; each claim cites a retrievable source. Release is deterministic: preflight is a rule engine, not a model. Where the customer workflow requires it, QRI can track external submission, witness, acceptance, rejection, and resubmission. Legal effect and discharge conditions remain customer- and contract-specific. Lifecycle: draft → review → approval → QMS return → optional external acceptance.
Scope honesty: the public interactive demo shows workflow UI. PoC backend and customer integration are engagement-scoped. Items under Roadmap in Deployment are not current.
Core concepts
- Source case: the QMS record a report is written from. Reports are normally created from an imported case; manual creation is an audited exception.
- Source-locked field: a critical value (part number, serial/lot, firmware revision, test temperature) bound to a single authoritative source. Conflicting sources block confirmation until resolved.
- Claim ledger: the compiled draft as atomic claims, each carrying its citation and an inference-run ID. The run record stores model registry ID, version, weights hash, prompt/template version, retrieval configuration, and generation parameters. Claims do not duplicate the full weights hash.
- Preflight: deterministic release gates. Blocking rules stop submission; reviewer warnings do not. Every failed row names its owner role and whether retry is safe.
- Role separation: authors hold EDIT_RECORD, reviewers hold APPROVE_REPORT; the author of a version can never approve it. The demo records a simulated approval decision. Production attribution and legal effect depend on customer-approved identity, signature, retention, and access controls.
- GQAR lifecycle: for government contracts, submit package → optional witness of RCA/corrective action → acceptance or rejection, per 방위사업 품질관리 규정 별지 제18호. Configurable for NATO-pattern AQAP 2110 reporting requirements when the customer workflow requires it.
- Model qualification: no model serves a production task until it passes measured gates (extraction accuracy, retrieval nDCG, unsupported-claim escape ≤1%, terminology accuracy), whether Modern Tech qualified or customer-approved. Each generation is an inference run linked from claims by inference-run ID.
Screen reference
- Work: Dashboard (open / blocked-by-preflight / ready / past-due, attention list); My Work; Quality Report Work Queue (shared register with saved views, filters, columns).
- Quality Records (QRI-native): Imported Cases (QMS/file ingestion airlock); Evidence Library (immutable evidence with exact provenance).
- Connected QMS (external): Quality Events, CAPA, Actions, Effectiveness, Suppliers, Document Register as read-mirrors, bannered EXTERNAL.
- Knowledge & Standards: Historical Knowledge (hybrid retrieval; similarity is never presented as proof of cause); Templates & Rules (8 types, versioned, clause citations); Terminology (controlled KO/EN register).
- Report form (11 tabs): Overview · Report Sections · Source Case & Sync · Confirmed Fields · Evidence Register · Historical Cases · RCA Assistant · Report Studio · Preflight · Review & Return · Audit Trail.
- Administration: AI Models & Runtime; Import & Onboarding; Workflow Configuration; Users & Access; System Validation (installation, operational, performance, security, and backup/recovery verification for the defense deployment profile; GAMP-5 is a future life-sciences packaging option, not the default defense story); System Audit; Connected QMS Modules.
Report types
Eight templates ship in the Modern Tech demo pack. Versions are demo configurations, not customer-validated production releases. Production use requires customer validation. Report output language is a template property (KDS templates → Korean; AS/AIAG → English), independent of interface language. Technical identifiers are never translated.
- NCR/CAR demo v3.2KDS 0050-9000 8.7 · 10.2 · KS Q ISO 9001
- 8D demo v2.1AIAG 8D methodology · IATF / customer-specific corrective-action requirements
- FRACAS demo v1.4MIL-HDBK-2155 · KDS 0050-9000 10.2 · 10.3.1
- GQA-DEFECT demo v1.0KDS 0050-9000 8.5.5.1 · 별지 제18호 · AQAP 2110 (configurable)
- COUNTERFEIT demo v1.0KDS 0050-9000 8.1.4 · 8.1.5 · AS5553 · AS6174
- QA-PLAN demo v1.0KDS 0050-9000 8.1.2 · 8.7.3 · 별지 제18호
- FAI/AS9102 demo v1.0AS9102C · AS9100D 8.5.1.3
- PPAP demo v1.0AIAG PPAP 4th Ed. · IATF 16949 8.3.4.4
Deployment & security
On-premise, air-gapped: no external API dependency. Model- and runtime-neutral: Modern Tech qualified local models on customer GPU, or a customer-approved internal AI platform. Claim provenance, preflight, approval separation, and QMS return stay with QRI. RBAC with change-controlled accounts (deactivated, never deleted). Append-only audit trails at record and system level.
E-signatures carry meaning, credential, and acknowledgement. Classification marking schemes are tenant configuration (e.g. US CUI), stamped on exports, never hardcoded.
Connector failure paths (offline / stale / read-only / rejected payload) fail safe: nothing partially written; retry safety is stated.
Roadmap (not current): HWPX/DOCX/PDF-A export, OCR/HWP/XLSX/instrument-CSV/photo ingestion, redlines and version compare, submission tracking, reporting analytics, customer-specific template packs. Do not treat these as demo-available.
Demo guide
Interactive frontend prototype with synthetic data throughout, watermarked. Not an approved quality record. One-click role entry (Quality Engineer / Quality Reviewer / System Administrator), no credentials. Backend PoC and live QMS integration are engagement-scoped, not implied by the UI alone.
Suggested path: open RPT-2026-0041 → resolve the firmware source conflict → compare precedents → record RCA (5-Why/fishbone) → compile the draft → repair claims → run preflight → submit → switch to Reviewer → approve with e-signature → return to QMS → record GQAR acceptance.
Failure demos: set the connector offline/stale; try to approve as the author (refused); try to assign an unqualified model (refused). Reset Demo restores the seeded state.
Glossary
Controlled Korean / English pairs
- 부적합
- nonconformance
- 시정조치
- corrective action
- 사전검증
- preflight
- 출처 잠금
- source-locked
- 특채
- concession
- 처분 결정 권한
- disposition authority
- 정부품질보증 대표자
- GQAR
- 입회
- witness
- 인수
- acceptance
- 반송
- rejection/return
- 발행
- release/issue
- 게시
- publish
- 유출점
- escape point
- 적격성 평가 qualification.
FAQ
What is QRI?
QRI (Quality Report Intelligence) is on-premise software for defense and regulated manufacturers that writes customer-required quality documents from the manufacturer's own registered evidence: nonconformance and corrective-action reports (NCR/CAR), 8D problem-solving reports, FRACAS failure reports, first-article inspection reports (AS9102), and PPAP packages. Every released factual claim traces to a retrievable source, and a deterministic preflight check gates release.
How does QRI prevent AI hallucination in quality reports?
QRI separates generation from release. Language models draft; a deterministic rule engine blocks release of any claim that lacks a cited source, any part number without an authorized source, and any causal conclusion without a verification requirement. The measured escape rate of ungrounded assertions into approved reports is the gating metric, reported with numerator, denominator, and method.
Does QRI replace our QMS?
No. QRI connects to the existing QMS (quality events, CAPA, suppliers, document register remain the system of record), imports cases, prepares and validates the report package, and returns it. Where required by the customer workflow, QRI can also track external submission, witness activity, acceptance, rejection, and resubmission. Legal effect and discharge conditions remain customer- and contract-specific.
Can QRI run in an air-gapped network?
Yes. QRI runs on-premise with no external API requirement. It is model- and runtime-neutral: use a Modern Tech qualified local model (typically 7B-32B class) or a customer-approved internal AI platform. Either path keeps claim provenance, preflight, and QMS return. A model must pass measured qualification gates before production use.
Which standards does QRI support?
Report templates cite governing clauses: KDS 0050-9000 (Korean defense QMS), KS Q ISO 9001, AS9102C first-article inspection, AIAG PPAP 4th edition, AIAG 8D methodology with IATF/customer-specific corrective-action requirements, IATF 16949, MIL-HDBK-2155 FRACAS, AS5553/AS6174 for counterfeit parts, and configurable AQAP 2110-pattern government quality assurance where required.
Is QRI available in Korean?
Yes. The interface, controlled terminology register (approved Korean/English quality-term pairs), and report output are fully bilingual. Report output language is set by the template, independent of interface language.