Skip to main content
Back to Projects
NLP · Transformers · Policy AnalysisImplemented

NLP — Transformer-Based Policy & Regulatory Intelligence

A transformer-based document intelligence stack that converts unstructured policy contracts and regulatory filings into a structured, searchable clause library — enabling underwriters and compliance analysts to query across large document portfolios.

BERT + GPT fine-tuned

Models

NER · clause extraction

Tasks

Elasticsearch

Index

Problem & use case

Underwriters and compliance teams manually reviewed policy wordings and regulatory filings to identify coverage gaps, endorsement conflicts, and regulatory obligations — slow and inconsistent at portfolio scale.

Solution & engineering approach

Domain fine-tuned BERT for token-level clause extraction combined with GPT-based clause normalization, indexed in Elasticsearch to support structured search and analytical querying across policy and regulatory document libraries.

Architecture & data flow

OCR &layoutExtractclausesNormalizeSearchindex
  1. 01

    OCR + layout parsing of policy / claim PDFs

  2. 02

    Domain fine-tuned BERT extracts entities and clauses

  3. 03

    GPT normalizes clauses to canonical schema

  4. 04

    Index structured output in Elasticsearch

  5. 05

    Expose query API for downstream analytics

Technology stack

TransformersBERTspaCyHugging FaceElasticsearch

Implementation status

Implemented — domain fine-tuned transformer pipeline with Elasticsearch query API

All ProjectsNext: RAG Pipeline — Insurance Documents