Deviation Investigation Drafting CAPA
Drafts CAPA root-cause analysis from your historical deviation database. QA engineers review and sign — never write from scratch.
PHARMA
For Quality, Manufacturing, Regulatory Affairs
AI that lives inside your GMP environment.
Fine-tuned on your SOPs, batch records, and deviation history. Runs on your hardware. DPDPA + Schedule M + 21 CFR Part 11 aligned from Day 1.
Data Stays On-Prem
ALCOA+ Audit Trail
IQ/OQ/PQ Shipped
Pilot Pass-Bar Guarantee
pharma-plant.svg — on-prem deployment
Built by Rishi Tode
· 15 years in semiconductors · Founder & CEO, Reyatech Systems Private Limited
· LinkedIn ↗
§2 · Use cases
Six workflows. All drafted by AI. All reviewed and signed by your team.
Pre-built drafting and review assistants for the work your QA, regulatory affairs, and manufacturing teams already do.
Batch Record Review MBR · EBR
Flags anomalies in manufacturing batch records before QA sign-off. Cuts review time by 40–60% based on internal pilots.
SOP Question-Answering CITATIONS
Any QC analyst, operator, or supervisor asks a question about an SOP in natural language. Answers come back with citations to the source SOP, version, and section.
Regulatory Filing Assistance CTD · eCTD
Drafts CTD / eCTD sections, stability summaries, and ANDA submissions from your structured study data. Regulatory affairs team reviews and signs.
Audit Preparation WHO · FDA · CDSCO
Reviews documentation completeness against WHO / FDA / CDSCO audit checklists. Flags gaps two weeks before the auditor walks in.
Change Control Assessment REGULATORY IMPACT
Evaluates the regulatory impact of a proposed manufacturing change against your change-control SOP and recent regulatory guidance.
§3 · ALCOA+ Compliance
The nine ALCOA+ attributes, mapped to evidence artefacts.
For every attribute, the control that satisfies it and the artefact you'll hand to your auditor. Nothing aspirational; nothing handwaved.
| Attribute | How OwnAI Satisfies It | Evidence Artefact |
|---|---|---|
Attributable |
Keycloak SSO + per-prompt user identity in Langfuse log. | user-attributed-prompt-response-export.csv |
Legible |
Native UTF-8 markdown rendering + immutable PDF export. | sample-exported-log.pdf |
Contemporaneous |
Server-time timestamps at request entry; NTP-disciplined clock. | ntp-sync-report.txt |
Original |
Raw prompt + raw response logged; no post-hoc editing path exists. | hash-anchored-log-entry.jsonl |
Accurate |
Customer-defined eval rubric run on every release. | eval-report-v{n}.pdf |
Complete |
All requests logged — including aborted and errored runs. | log-completeness-audit.pdf |
Consistent |
Frozen production model + adapter version pinning. | model-registry-export.json |
Enduring |
Restic encrypted backups with tested restore (RPO 24h). | backup-restore-test-report.pdf |
Available |
On-premises; available to your auditors without delay. | audit-walk-in-runbook.pdf |
§4 · Regulatory mapping
Four frameworks. Four explicit answers.
Expand each card for the per-clause posture your compliance head will ask you to defend.
DPDPA 2023 + DPDP Rules 2025Data Protection · India
Reyatech operates as Data Processor under §2(p). All processing occurs on your hardware. No cross-border transfer. Phase 3 obligations bind 13 May 2027 — we are aligned today.
DPA template provided per engagement. We do not become Joint Fiduciary under any circumstance — using your data to improve another customer's model is contractually prohibited.
Schedule M (revised 2024) / ALCOA+Indian GMP · CDSCO
Immutable prompt/response logs via Langfuse (on-prem). Frozen production models with version-locked adapters. See ALCOA+ compliance matrix above for per-attribute mapping.
Data-integrity controls match the revised 2024 Schedule M elevation of pharmaceutical data governance — system clock NTP-disciplined, log entries hash-anchored, no edit path.
21 CFR Part 11US FDA · Electronic Records
Keycloak access control + audit-trail integrity + electronic signatures (when required) + system validation. IQ/OQ/PQ artefacts are included in every engagement as named deliverables, not extras.
Role-based access at the model, dataset, and conversation level. MFA mandatory for any role with model-refresh privileges.
GAMP 5 / CSVComputer System Validation
Deployed as a frozen, version-controlled system. No self-learning drift in production. The validation lifecycle aligns with GAMP 5 Category 4 (configured product) handling.
Each model refresh is treated as a change-control event with associated re-validation activities. URS / FS / DS templates ship with the engagement.
§5 · Comparison
How OwnAI is different from cloud LLM seats.
The same eight questions your CISO will ask, and the answers from each side.
| Capability | ChatGPT Enterprise / Gemini / Copilot | OwnAI |
|---|---|---|
| Data location | US / EU cloud | Your server room |
| DPDPA Phase-3 posture | DPA required + cross-border transfer review | Native — data never leaves India |
| Audit trail | Limited API logs | Full Langfuse log, on-prem, hash-anchored |
| IQ/OQ/PQ artefacts | Not provided | Provided per engagement |
| Cost model | Per-seat, forever | One-time + AMC at 18–22% |
| Model customisation | System prompts only | Full LoRA fine-tune on your corpus |
| Schedule M alignment | Cannot satisfy contemporaneity | NTP-disciplined log, immutable entries |
| Auditor walk-in | Not feasible | Walk in; we'll meet you there |
§6 · ROI · Pharma scenario
A typical pharma QA + regulatory team, costed honestly.
Default: 30 QA engineers + 10 regulatory affairs analysts. ChatGPT Enterprise baseline. 5-year horizon. Substitute your own numbers in the calculator.
All-in cost: setup fee + hardware + 5-year AMC, pre-GST. Cloud baseline assumes flat ChatGPT Enterprise pricing at current INR ₹ exchange.
Scenario · 5 years
Pharma vertical
QA engineers
30
Regulatory affairs
10
Hardware tier
L40S Server
Pilot
4 weeks
5-year savings
₹1.05 Cr · 74%
§7 · Pilot pass-bar & measured outcomes
Pilot pass-bar — what we will refund if we miss.
These are the eval-rubric thresholds we sign in the pilot SOW before kickoff. Miss any of them at the 4-week pilot review and you pay zero.
Measured in production — pharma case study.
Numbers from a named (under NDA) customer 60 days post go-live, against a signed eval rubric. Different from the pilot targets above: these are what one production deployment actually delivered.
Top-10 Indian pharma · 12,000 employees · multi-site · measured · signed eval rubric · real customer
- −60%deviation drafting time
- 0data-residency exceptions
- 7-monthbreak-even
- 93%SOP Q&A accuracy
Read the full case study → Request a reference call
Customer name withheld under NDA. Real metrics, signed eval rubric pre-kickoff.
§8 · Pharma pledges
Four things we will never do — on a pharma engagement.
These are clauses, not slogans. They live in §6 of every pharma SOW and are countersigned at pilot kickoff.
We never use your batch records or SOPs to train any other customer's model.
We never deploy a self-learning production model. Production is frozen until you sign off on a refresh.
We never ship a production model without a validation cycle aligned to GAMP 5 Category 4.
We never make changes to the production system without a change control record.
Next step
Bring your top three deviation reports.
A 30-minute call with someone who has read Schedule M (2024). Bring three recent deviation reports — anonymised is fine — and we'll walk through how the drafting assistant would have handled each one.