International Journal of Scientific Research and Engineering Development

The rapid growth of unstructured scientific literature presents significant challenges for automated knowledge acquisition and semantic querying. Existing pipelines frequently suffer from unreliable extraction, lack of traceability, and hallucinationprone retrieval. This paper proposes KnowPro, a unified, modular framework that integrates structure-aware document ingestion, hybrid symbolic-neural knowledge extraction, confidence-gated dual-layer storage, provenance-aware knowledge graph construction, and a strategy-based Graph Retrieval-Augmented Generation (Graph-RAG) architecture. The hybrid extraction engine combines rule-based Open Information Extraction patterns with a pre-trained SciBERT BIO token classifier, merging outputs through a deduplication-and-max-confidence mechanism. A configurable three-tier routing system partitions extracted triples by confidence into a high-fidelity Neo4j reasoning graph and a fully auditable PostgreSQL triple store. The Graph-RAG retrieval layer implements five deterministic traversal strategies—targeted, chained, variable-hop, shortest-path, and sharedneighbor—with n-gram and Levenshtein-based entity resolution and strict prompt-level grounding. Internal validation on a medical knowledge domain demonstrates end-to-end pipeline correctness across 105 automated tests and four representative endto-end traces. In this version, KnowPro is presented as an extensible architecture for trustworthy knowledge graph construction and structured semantic retrieval from scientific corpora, with several evaluation items still scoped as implementation validation rather than benchmark-level comparison.

Swapnil Kale, Sanchit Joshi, Aaradhya Kulkarni, Vardhan Bhanuwanshe, Prof. Varsha Kulkarni,"KnowPro: A Confidence-Aware Knowledge Graph Construction and Graph-RAG Retrieval Framework for Unstructured Scientific Text" International Journal of Scientific Research and Engineering Development, V9(3): Page(1267-1273) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.