JPMorgan AI Research · Finance
DocLLM-Finance
A layout-aware language model from JPMorgan designed for understanding complex financial documents with rich formatting and tabular data.
Overview
DocLLM-Finance extends large language models with spatial layout understanding, enabling them to process complex financial documents where the arrangement of text, tables, and figures carries meaning. Developed by JPMorgan's AI Research team, the model incorporates bounding box information and document structure into its reasoning, making it particularly effective at understanding financial statements, regulatory filings, and structured reports where traditional text-only models struggle.
Architecture
Layout-aware transformer with spatial encoding
Input
Text + bounding box coordinates
Document Types
Financial statements, forms, reports
Provider
JPMorgan AI Research
Availability
Research publication; limited access
Capabilities
Layout-aware document understanding
Financial table extraction and analysis
Multi-page document comprehension
Structured data extraction from unstructured documents
Form and invoice processing
Use Cases
Extracting data from complex financial statements and tables
Processing loan applications and supporting documentation
Automating regulatory reporting from formatted documents
Analyzing multi-page investment prospectuses
Pros
- +Understands document layout critical for financial documents
- +Handles complex tables and multi-column formats
- +Backed by JPMorgan's deep financial domain expertise
- +Addresses a key gap in financial document AI processing
Cons
- -Not publicly available or open-source
- -Limited to document-level tasks; not a general financial LLM
- -Requires document pre-processing with OCR and layout detection
- -Research-stage without clear production deployment pathway
Pricing
Not publicly available. Research-stage model from JPMorgan. Enterprise solutions may be integrated into JPMorgan's internal systems.