Chemistry, Manufacturing and Controls (CMC) process development generates vast technical information fragmented across documents. The authors built a platform that ingests these documents into a dual-layer : a base lexical layer organizing documents into a hierarchy, and an intelligence layer extracting ontology-aligned entities and cross-document relationships. LLM agents select the appropriate retrieval path for each question. On a benchmark of 505 questions from 38 Sanofi development reports, the lexical layer achieved 95% multiple-choice accuracy (Tier 1) and an 85% pass rate under a stricter LLM-judge evaluation (Tier 2), with performance dropping on comparative and corpus-wide questions. This work demonstrates a structured approach to transforming fragmented document repositories into queryable process intelligence, addressing a critical knowledge-management challenge in the pharmaceutical industry.
Dual-layer knowledge graph connects fragmented pharmaceutical process-development documents
The system combines document-level retrieval with ontology-based links, achieving 95% multiple-choice accuracy and an 85% LLM-judge pass rate on proprietary reports.
Industry
Reza Amirmoshiri · Faryad Sahneh · Yasser Jangjou
Sanofi US
Research Digest··2 min read
In one line
A dual-layer knowledge graph with LLM agents retrieves answers from CMC documents with 95% accuracy on a 505-question benchmark.
Before you read this
Two ideas this paper builds onRead it as
Amirmoshiri, Sahneh and Jangjou built an agentic platform that transforms heterogeneous Chemistry, Manufacturing and Controls documents into a queryable, provenance-linked knowledge graph.
The authors created a dual-layer for pharmaceutical process development.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.