Chemistry, Manufacturing and Controls (CMC) process development generates massive technical information across the continuum from drug discovery to commercial manufacturing. This information is traditionally scattered across functions and stored in heterogeneous formats, creating traceability gaps and raising knowledge‑management costs during technology transfer and regulatory filing. We present a modular agentic‑AI platform that transforms a heterogeneous corpus of process‑development documents into a queryable dual‑layer knowledge graph. The base layer ingests digital, scanned, handwritten, and multilingual documents without loss, building a lexical graph organized as Document‑Section‑Chunk hierarchy. The intelligence layer extracts ontology‑aligned entities and links cross‑document concepts through a provenance‑anchored domain graph. LLM agents operate across both layers, selecting the retrieval path best suited to each query. A novel three‑tier evaluation protocol measures the deployment fidelity of a retrieval‑augmented generation (RAG) system on proprietary data, using 505 questions curated from 38 development reports of a small‑molecule program. Tier‑1 multiple‑choice accuracy reaches 95%, indicating strong platform reliability; Tier‑2 LLM‑judge pass rate of 85%—which drops on comparative and corpus‑wide questions—reveals a failure taxonomy that Tier‑1 accuracy alone misses. A router agent switches between layers according to question type. We anticipate this protocol will enable future designers of agentic platforms to assess their systems on non‑public databases, and that graph‑based architectures will see broader adoption in pharma as a means to convert fragmented document repositories into structured process intelligence.
Review