Bio-IT Regulatory Platform · Songdo, Incheon

Your IND filing
has gaps you
haven't found yet.

AI BioAudit audits your CMC manufacturing documentation before regulators do — identifying data integrity failures, OOS events, and traceability gaps that manual review consistently misses.

Request Early Access See a live audit demo →
Audit Report · AV-101
Scanning
→ Ingesting stability_report_batch004.pdf
→ Parsing table structure (nested)...
→ Running ICH Q1A(R2) logic gates...
→ Evaluating temperature compliance...
⇂ CRITICAL: Thermal excursion Month 2 (−72.4°C)
→ Running potency trend regression...
⇂ MAJOR: 40% potency loss — OOS projected Month 8.5
→ Checking purity thresholds...
⇂ OOS: Month 6 purity 94.5% < 95% spec limit
✓ Audit complete · 3 flags require attention.
$1B+
Market cap destroyed in the Invossa documentation failure
15yr
Cell identity errors went undetected in Invossa CMC records
600K+
Litres biomanufacturing capacity concentrated in Songdo
The Stakes

The cost of one
undiscovered error.

"15 years of undetected cell identity errors were a failure of traceability, not science."

Invossa Case · South Korea · 2019

The 2019 Invossa license revocation exposed a systemic failure in CMC documentation integrity that affects Korean biotechs across the sector — a widening gap between world-class science and the forensic data requirements of global regulators. Korean biotech companies face increased risk of Clinical Holds during FDA/EMA filings due to structural gaps in how manufacturing data is prepared and presented.

$1B+
Market capitalization destroyed following the Invossa withdrawal
15yr
Duration of undetected cell identity errors in manufacturing records
Hold
U.S. FDA clinical trials halted pending documentation review
CMC
Module 3 data integrity — the most common regulatory failure point
Gap 01
Data Integrity
Korean biotech reports lack the forensic data lineage required by FDA and EMA. Manufacturing records describe outcomes but do not provide the traceable evidence chains regulators demand.
Gap 02
Global Standards
Domestic documentation follows Korean conventions. Global approval requires evidence-based dossiers aligned with ICH Q1/Q5 guidelines, 2026 FDA CBER reforms, and 21 CFR Part 11 standards.
Gap 03
Cultural Risk
Systemic suppression of technical deviations within corporate structures creates hidden regulatory risk that surfaces only during FDA review — when it is already too late and too costly to correct.
The Product

How AI BioAudit
works.

A three-module pipeline that ingests your manufacturing documents, applies codified regulatory logic, and surfaces the risks that manual review misses — before they reach an FDA or EMA submission.

01
📄
Layout-Aware Extraction
Preserves scientific notation and complex nested tables from unstructured PDFs — capturing exactly what generative AI distorts or loses. Your data, intact, structured, and ready for analysis.
02
⚙
Regulatory Logic Gates
ICH Q1/Q5 guidelines and 2026 FDA CBER reforms are codified as deterministic if/then rules — not inference. Biophysical anomalies are caught by logic, not by probability or plausible-sounding outputs.
03
⚑
Anomaly Detection & Scoring
Flags OOS results, data drifts, missing impact assessments, and statistical inconsistencies. Generates a prioritised Clinical Hold Risk Score — reproducible, explainable, and defensible under regulatory scrutiny.
Live Demo

Watch the engine find
what humans miss.

The simulation below audits a real-world AAV gene therapy stability report — the same documentation type that triggered the Invossa clinical hold. Three regulatory landmines. All caught automatically.

Ready to scan
BATCH-2026-004_Stability_6Month.pdf
MonthDateTemp (°C)Purity AUC%Potency (IU/mL)SterilityAI Flag
02025-10-01-81.298.5%1.2 × 10¹²Pass—
12025-11-01-79.598.1%1.1 × 10¹²Pass—
22025-12-01-72.497.2%9.8 × 10¹¹PassCRITICAL
32026-01-01-80.196.8%8.5 × 10¹¹PassMAJOR
62026-04-01-78.894.5%7.2 × 10¹¹PassOOS
⚠
Critical — ICH Q1A(R2) Violation · Temperature Excursion
Month 2 temperature excursion to −72.4°C. Report lacks mandatory impact assessment for potential capsid instability and sub-visible aggregate formation. ICH Q1A(R2) Annex 3 requires biophysical re-characterisation (aggregation/sub-visible particles) before IND submission. Simply recording the excursion without investigation is insufficient for FDA review.
↓
Major — Statistical Trend Failure · Potency Degradation
Potency regression 1.2→7.2 ×10¹¹ IU/mL represents a 40% loss over 6 months. Linear regression with 95% confidence interval projects out-of-spec expiry at Month 8.5. Manufacturer conclusion "No significant degradation" is mathematically unsupported. FDA requires quantitative trend analysis; qualitative characterisation is insufficient.
✗
OOS — Out-of-Specification · Month 6 Purity Below Threshold
Month 6 purity 94.5% falls below the ≥95% specification threshold for clinical-grade AAV materials. This is a formal Out-of-Specification event requiring investigation and documented root cause analysis. Report characterises this result as "acceptable" — this misclassification itself constitutes a data integrity failure.
Architecture

Beyond generative AI.
Deterministic logic.

01
Layout-Aware Document Extraction
Multimodal parsing preserves complex nested table structures, scientific notation, and cross-document data references — the exact information that large language models distort or silently discard.
02
Codified Regulatory Rule Engine
ICH Q1, Q5, and 2026 FDA CBER reforms encoded as deterministic if/then logic. Every flag includes an explicit regulatory citation. Audit results are reproducible across runs — a requirement for evidentiary use.
03
Bank-Grade Data Security
All biotech manufacturing data processed under enterprise-grade cloud isolation. Zero-retention API agreements. Full ISMS-P compliance — the Korean regulatory standard for biotech and healthcare data residency.
Why not just use ChatGPT?
Generative AI infers.
Regulatory auditing requires proof.
Large language models produce plausible-sounding outputs. That is exactly what you do not want when the output is used as evidence in an FDA submission. AI BioAudit's audit logic is deterministic: the same input always produces the same output, with an explicit regulatory rule citation for every flag. Reproducible. Explainable. Defensible before any regulator in the world.
Founding Team

Dual expertise in biotech
and systems architecture.

CK
Christoph A. Kahl
CEO & Founder
20+ years in Cell and Gene Therapy with direct hands-on involvement in IND and BLA regulatory filings. Industry leadership at Amgen, Vir Biotechnology, and Atara Bio. Ph.D.-level scientific expertise applied to the specific compliance gaps Korean biotechs face when targeting FDA and EMA approval — and a first-hand understanding of where documentation failures occur in practice.
Cell & Gene Therapy IND / BLA Filings CMC Module 3 Amgen · Vir · Atara Ph.D. Virology
RM
Ricardo Martinez
Co-Founder & Lead Developer
8+ years in AI systems engineering and technical product leadership. Former algorithms engineer at ZEISS Industrial Metrology, building 94%+ accuracy defect detection pipelines for Apple and Samsung — automated inspection architecture directly analogous to CMC audit systems. CEO of Morningstar Ventures LLC. Based in Songdo, Incheon for 4+ years as a permanent resident with active engagement in IFEZ programs for foreign entrepreneurs.
AI Systems Computer Vision ZEISS · Apple · Samsung Songdo 4yr IFEZ Network
600K+
Litres biomanufacturing capacity
100+
Biotech startups in Songdo cluster
4yr
Founding team embedded in Songdo

AI BioAudit is built inside the same bio cluster it serves — with active IFEZ relationships and direct access to the Incheon Technopark and K-Bio Lab Hub ecosystem. We are not entering this market from the outside.

Pilot Programme · 2026

We are onboarding
founding partners.

We are currently working with a small cohort of CGT developers in the Songdo and Incheon cluster to validate the audit engine against real manufacturing documentation. If you are preparing an IND or BLA filing and want to know what an automated audit surfaces in your CMC records — speak to the founding team directly.

Speak to the Team contact@aibioaudit.com →

Targeting 2–3 Songdo CGT pilot partners for Q3–Q4 2026. Pilot access is by direct engagement only.