DealSight
NLP pipeline for automated extraction, summarisation, and risk-flagging of M&A deal terms from unstructured financial documents.
Legal and financial analysts spend hours reading merger agreements to extract key terms — price adjustments, MAC clauses, termination fees, and closing conditions. DealSight automates this with a multi-stage NLP pipeline that ingests raw deal documents, extracts structured fields, summarises each clause, and flags non-standard or high-risk language.
A retrieval-augmented generation (RAG) layer allows analysts to query any deal in natural language — “What are the termination fee provisions?” — and receive cited, grounded answers from the source document.
Key contributions
- Named-entity recognition fine-tuned on financial agreements to extract deal parties, values, and dates.
- Clause-level risk scoring based on deviation from market-standard language.
- RAG interface with source citation for analyst Q&A.
Tech stack: Python · spaCy · Hugging Face Transformers · LangChain · FAISS · Streamlit