Image

Information Extraction

Rapidly build AI-powered applications that extract information from unstructured text, PDF, tables, or forms from millions of documents without expensive hand-labeling using Snorkel Flow.

Request demo
Image
Image

Top US Bank

99.1%

accuracy for contract intelligence built using Snorkel Flow in under 24 hours

Overview

All Data Extracted with No Context Lost

Extract useful data from any tables, cells, and forms linked to all headers, units, or references.

Image

Faster, Lower-cost Development

Use programmatic labeling to develop high-quality AI applications in hours instead of spending weeks or months on expensive hand-labeling.

Image

Rapidly Adaptable

Monitor for changes in the data, and rapidly adapt using built-in error analysis tools. Zoom in on errors to fine-tune training data & models with guided iteration.

Image

High-accuracy Models

Leverage large amounts of labeled and unlabeled data, NLP primitives, and state-of-the-art model architectures to build high-accuracy models.

Image

Flexible Integrations

Easily integrate labeling, training, and analysis pipelines defined over diverse input types–text, PDF, HTML, and more–with downstream applications using APIs or a Python SDK.

Use Cases

Information Extraction Customized for Your Workflow

FINANCIAL SERVICES

Market Analysis
Banks can extract information from financial reports to explore market trends.

TELECOM & CYBER

Network Optimization
Network operators can extract information device details from technical documents to optimize network resources.

HEALTHCARE

Drug Discovery
Pharmaceutical companies can automate data extraction from clinical trial records for digital pathology.

INSURANCE

Risk Evaluation
Insurance firms can evaluate risks associated with a policy by extracting contextual data from contracts and forms.

SOFTWARE

Invoice Processing
Software companies can extract information from invoices or receipts for accounting or expense analysis.

RETAIL

Product Catalogs
Online marketplaces can extract product attributes from tables, lists, and forms for cataloging.

FINANCIAL SERVICES

Smart Search
Banks can create smart search indexes to locate records with specific attributes, like loan rates or credit scores.

Case Study

Top US Bank

A top U.S. bank uses Snorkel Flow to quickly build AI applications that classify and extract information from contracts and other legal documents.

Problem

The bank estimated that, for a time-sensitive use case, labeling data by hand would take over a month.

Solution

With Snorkel Flow, the team produced a AI-powered contract intelligence application that was over 99% accurate in under 24 hours.

Result

The resulting AI application was quickly and easily adapted to new problems.

<24 HOURS

to develop first model

99.1%

model accuracy

250K

documents labeled

How Snorkel Flow Works

Build Flexible AI Apps for Extraction

Build flexible extraction applications that preserve structural and tabular information and generalize beyond brittle rules. Eliminate hundreds of hours of manual labor by programmatically labeling data with powerful labeling functions. Train state-of-the-art models and analyze performance using intuitive tools. Chain ML tasks together and deploy your extraction application as an API or Python SDK with one click.

An End-to-end ML Platform

Designed for Collaboration

Image

Data Scientist Friendly

  • Integrated Jupyter notebooks
  • Instant analysis tools
  • Ready-to-use models
Image

Domain Expert Friendly

  • Intuitive, no-code UI
  • Rich dashboards and visualizations
  • Full-featured, push-button error analysis
Image

Developer Friendly

  • Platform access via Python SDK
  • Online or batch API deployment
  • Containerized software for cloud or on-premises deployments

Research

Based on Years of Novel Research
Learn more about groundbreaking techniques for machine learning and weak supervision developed by the Snorkel AI team at Stanford AI Lab and beyond.

RESEARCH

Training Complex Models with Multi-Task Weak Supervision

A. Ratner, et al, AAAI 2019

RESEARCH

The Role of Massively Multi-Task and Weak Supervision in Software 2.0

A. Ratner, et al, CIDR 2019

RESEARCH

Software 2.0 and Snorkel: Going Beyond Hand-Labeled Data

C. Ré, KDD 2018 (invited)

RESEARCH

Snorkel: Fast Training Set Generation for Information Extraction

A. Ratner, et al, SIGMOD 2017

BLOG

Snorkel AI:
Putting Data First in ML Development

Alex Ratner

RESEARCH

Learning to Compose Domain-Specific Transformations for Data Augmentation

A. Ratner, et al, NeurIPS 2017

RESEARCH

Learning the Structure of Generative Models without Labeled Data

S. Bach, et al, ICML 2017

RESEARCH

Fonduer: Knowledge Base Construction from Richly Formatted Data

S. Wu, et al, SIGMOD 2018

RESEARCH

Deep Text Mining of Instagram Data without Strong Supervision

K. Hammar, et al, ICWI 2018

RESEARCH

Data Programming: Creating Large Training Sets, Quickly

A. Ratner, et al. NeurIPS 2016

RESEARCH

A Machine-compiled Database of Genome-wide Association Studies

V. Kuleshov, et al. Nature 2019