AI-Driven CTD Regulatory Dossier Automation System

Our main objective was to take the friction out of pharmaceutical regulatory filings by building a dependable, AI-powered compliance system that handles the heavy lifting of CTD dossier assembly:

• Smarter Drafting: Use Retrieval-Augmented Generation (RAG) and structured templates to generate narrative sections (SPC, RMS, PSUR) grounded strictly in verified source data.

• Automated CTD Splitting: Automatically break down massive master files into clean, organized sections for non-clinical and clinical reports (Modules 4 & 5).

• Reliable Literature Verification: Check in-text citations across PubMed, Crossref, and Google Scholar simultaneously, using an LLM to lock down accurate, audit-ready references.

• Audit-Proof Traceability: Track the origin of every single paragraph with confidence scores to satisfy strict regulatory inspections.

• Real-World Impact: Shrink filing timelines from five to six months down to two to three weeks, removing costly human errors along the way.

Rolling out this platform completely changed how our regulatory team works day to day. The most immediate win was time: preparing an entire registration dossier used to tie up senior specialists for five to six months, but we brought that down to just two to three weeks. That alone helped us get products to target international markets months ahead of schedule.

The citation engine solved one of our biggest administrative headaches by hitting a 92% automated match rate across PubMed, Crossref, and Google Scholar, saving countless hours of manual literature checks. Core documents like the 12-section SPC now come together in about seven and a half minutes instead of days of drafting. Most importantly, we didn't sacrifice compliance for speed. Every generated passage carries full source attribution and a confidence score, giving our teams the audit-ready transparency they need to submit with total confidence

We engineered a high-performance stack built specifically for speed, precision, and enterprise data privacy:

• Core AI & Models: We paired Groq LPUs for rapid text inference with Azure OpenAI for enterprise-grade tasks. Document embeddings run locally via Ollama using nomic-embed-text, keeping sensitive data on-premise.

• Retrieval & Caching: ChromaDB manages product-isolated collections via fast HNSW indexing, while Redis 7 slashes redundant API costs through instant prompt caching and state management.

• Citation & Document Handling: The engine queries PubMed, Crossref, and Google Scholar in parallel, with Chandra OCR extracting complex scanned reports and python-docx handling DOCX-to-PDF formatting.

• Application & Setup: A responsive React 18 frontend connects to a lightweight FastAPI backend over 18 dedicated endpoints, containerized with Docker Compose on an Intel Core i9-13900K rig with 64 GB RAM

The biggest hurdle wasn't just connecting APIs; it was engineering an AI pipeline that could meet unforgiving pharmaceutical standards where hallucinations aren't an option.

First, we had to process heterogeneous, complex layouts from dense tables to scanned clinical reports using OCR and custom chunking to ensure source text stayed semantically intact. Second, managing multi-hundred-page master files required robust regex engines to slice documents cleanly into standardized CTD folder structures without corrupting non-clinical data.

The most intricate engineering challenge was the citation engine: coordinating asynchronous queries across three external scholarly databases simultaneously, resolving domain synonyms, and scoring candidates with an LLM in real time to reach a 92% match rate. Finally, building end-to-end source provenance and Redis caching ensured the system stayed fast, cost-efficient, and fully audit-ready for regulatory inspections.

In the pharmaceutical sector, getting products registered across international markets is often held hostage by paperwork. Compiling a Common Technical Document (CTD) dossier manually used to take our regulatory team five to six exhausting months per product. Experts spent endless days copying data across templates, verifying hundreds of medical citations by hand, and cross-checking documents for discrepancies. To eliminate this bottleneck, Herbion Pakistan teamed up with FAST-NUCES to engineer an AI platform that turns months of administrative friction into a matter of weeks.

Rather than relying on generic AI that risks making things up, we built a domain-specific, auditable architecture. The system uses Retrieval-Augmented Generation (RAG) to draft critical narrative sections like the Summary of Product Characteristics (SPC) grounded strictly in verified internal source files. For extensive clinical and non-clinical reports (Modules 4 & 5), the platform automatically breaks down massive master files into standardized CTD folders. Crucially, its citation engine searches PubMed, Crossref, and Google Scholar in parallel, expanding medical terms and using an LLM validator to verify references with 92% accuracy.

Because regulatory authorities demand complete transparency, the platform logs the exact source passage, similarity metric, and confidence score for every single paragraph generated. The business impact is immediate and practical: dossier turnaround has dropped from nearly half a year to just two to three weeks. This drastic reduction cuts operational overhead, eliminates manual transcription errors, and significantly accelerates product time-to-market across export territories.

The project was conceptualized, engineered, and deployed at Herbion Limited's corporate headquarters in Karachi, Pakistan, developed in collaborative partnership with FAST-NUCES Karachi.

While built and operated domestically, the system is designed with an expansive global reach to drive commercial expansion across diverse international territories. The platform automates and standardizes regulatory Common Technical Document (CTD) submissions, adapting seamlessly to the distinct health authority guidelines and filing requirements of varied global export markets.

By serving as a centralized compliance hub, the platform governs cross-border registration workflows worldwide. It ensures end-to-end alignment across overseas regulatory jurisdictions, transforming heterogeneous technical documentation into standardized, globally accepted dossier structures.