Precision document retrieval in oncology drug development.

Aclinical-stage oncology biotech with 210 employees is running two Phase 2 drug candidates simultaneously. The regulatory affairs team manages a dense and growing corpus of documents — Investigational New Drug applications, clinical protocols, safety narratives, FDA meeting minutes, and correspondence accumulated across years of trial activity.

Every claim a regulatory affairs team makes about a drug’s safety profile, dosing rationale, or trial design must be traceable to the exact source document and section it came from. That is not a preference — it is a requirement of the work. And it is exactly what the tools the team had available could not reliably do.

The company managed its regulatory document library in SharePoint. When Microsoft Copilot Studio with SharePoint Agents became available, the team evaluated it — it is the right starting point, and for many organizations it is sufficient. For this client, it was not.

Copilot synthesizes across documents to produce a single answer. In most business contexts, that is exactly what you want. In regulatory affairs, it is a liability. A synthesized answer that blends language from two different documents — or that paraphrases rather than reproduces the exact approved wording — is not usable. Regulatory staff submitting responses to the FDA cannot cite a plausible-sounding answer. They need the precise language, from the precise section, of the precise document it came from. Copilot could not reliably provide that.

The standard industry alternative is Veeva Vault — a purpose-built regulatory document management platform with AI search capabilities. The problem is that Veeva requires documents to be uploaded to its own cloud environment. This client operates under strict data residency requirements. Sensitive regulatory documents — IND applications, safety narratives, FDA correspondence — could not leave their AWS infrastructure.

They needed precision retrieval with verified citations. Built to run entirely within their own environment.

Coral Tree deployed Chronicle — its proprietary document intelligence system — adapted to run natively within the client’s AWS environment. No data leaves the client’s infrastructure at any point in the pipeline.

The first phase was a security and data residency review, conducted before any technical build began. Once the architecture was validated against the client’s compliance requirements, Chronicle’s ingestion pipeline was configured for the specific taxonomy of regulatory documents: INDs, clinical protocols, safety narratives, FDA meeting minutes, and formal correspondence.

Each document type has distinct structural conventions — section numbering, header patterns, amendment hierarchies — and the ingestion pipeline was tuned to preserve that structure through indexing, so that citations can be traced not just to a document but to a specific section within it.

When a regulatory affairs staff member submits a plain-English query — “What does our IND say about dose modification criteria for Grade 3 adverse events?” — Chronicle searches the full indexed corpus and returns the specific passages that directly address the question. Each passage is returned with an exact citation: document name, section number, version, and page reference.

The system does not synthesize or paraphrase. It retrieves and attributes. The regulatory affairs officer reads the returned passages, sees precisely where each one came from, and uses them to draft the FDA response with full confidence in the sourcing.

This distinction — retrieval with citation versus synthesis without it — is what makes Chronicle appropriate for this context and Copilot insufficient. The staff member’s job is regulatory judgment. Chronicle’s job is to surface the right source material, accurately and completely, so that judgment can be applied to verified information.

Chronicle surfaces confidence signals alongside each result — flagging when a query falls outside the indexed corpus or when retrieved passages are weakly matched to the query. Regulatory staff are never left uncertain about whether the system has fully covered the relevant document space. If the answer is not in the corpus, Chronicle says so.

The citation accuracy figure is the foundation of everything else. A retrieval system that regulatory affairs staff cannot trust produces no time saving — every result still has to be manually verified against source documents. At 98% accuracy, the system earns the team’s reliance. The 1-to-2-hour manual search process becomes a 5-minute query.

The company did not have to choose between AI-powered document intelligence and data residency. Chronicle runs inside their infrastructure, under their security controls, with no third-party cloud dependency. The regulatory corpus stays where it belongs — and the team that depends on it can finally search it at the speed the work requires.