Semantic Link Finder Across Banking Statements and Contracts

A single banking relationship generates statements, contracts, and reports that all mention overlapping accounts, parties, and terms, yet each file is read in isolation. Answering a question like which agreements reference this fee structure means manually cross-reading documents, which is exactly the tedium this project removes. The intern builds a python application with a streamlit interface where users upload multiple banking documents. Each document is chunked and embedded into a vector database, allowing the system to surface passages from different files that discuss the same entities or topics. A retrieval-augmented generation pipeline answers user questions about connections between documents, and generative AI writes clear explanations of each relationship it finds, for example noting that a clause in one contract modifies a term defined in another. The interface lets users explore these links interactively, moving from a question to the exact supporting passages. Through the build, the intern learns cross-document retrieval, entity-centric exploration, and grounded explanation generation with RAG, gaining a practical feel for how AI turns a folder of disconnected files into a navigable body of knowledge.