Conversational AI

Build high-quality NLP and conversational AI applications by training state-of-the-art models with your data using Snorkel Flow.

Request demo



Case Study —

IBM Bootstrapped Chatbots with Weak Supervision

IBM Research used a weak supervision-based framework to develop a novel search, label, and propagate (SLP) architecture to bootstrap intent classification using existing chat logs.

More labels generated vs. hand-labeling
To develop the first custom ML model
Increase in model accuracy vs. hand-labeling
Accuracy for contract classification
Increase in precision vs. hand-labeling
Contracts processed in minutes

Read case studies

Deliver Engaging Experiences

With Snorkel Flow, you can build powerful conversational AI capabilities to understand customer intent, model topics, and analyze sentiment at an utterance or conversation level. Achieve higher quality with purpose-built models trained on your data–no costly hand-labeling needed.

Customer: I need to transfer 1250 dollars [TRANSFER_MONEY]
Agent: Which account do you want to withdraw funds from?
Customer: Checking [FROM_ACCOUNT]
Agent: Please confirm your 4-digit passcode
Customer: 1132 [PASS_CODE]

High-Accuracy Models

Train state-of-the-art NLP models from Snorkel Flow’s built-in model zoo, or custom models via Python SDK with push-button UI. Deploy language models that take advantage of all previous utterances.

Faster Development

Accelerate development with template-driven visual builders for many common conversational use cases and a wide range of labeling functions and NLP tools.

Is the tone of this conversation positive or negative?

Did the agent schedule a follow-up with the customer?

What services did the customer request during this interaction?


Adaptable Applications

Monitor performance drifts in Labeling Functions or the purpose-built models. Rapidly adapt to changes in data or business objectives without relabeling from scratch.


Collaborative Workflows

Put customer service agents' or other customer experience experts' knowledge to use with a no-code, push-button UI to create labeling functions.


Why Snorkel Flow —

A Radically New Approach to AI

Conventional AI approaches rely on generic third-party models, or brittle rule-based systems, or armies of human labelers. With Snorkel Flow, programmatically labeling unlocks a new workflow that accelerates AI app development.

With Snorkel Flow

  • Customize state-of-the-art models by training with your data & adapt to changing data or goals with a few lines of code.
  • Leverage cutting-edge ML to go beyond simple rules, and retain the flexibility to audit and adapt.
  • Label thousands of data points programmatically in hours, while keeping your data in-house and private.

With Conventional Approaches

  • Pre-trained vendor models often don’t work on your data, no way to customize, adapt, or audit.
  • Rules-based approaches often don’t perform well on complex data or adapt easily to changes in data or goals.
  • Hand-labeled ML is hugely expensive, with often no way to iterate and adapt, be privacy compliant, audit, or reuse.

How Snorkel Flow Works —

An End-to-end ML Platform Centered Around Data

Designed to accelerate AI application development by removing the training data bottleneck.



Label & Build
Label and build training data programmatically in hours without months of hand-labeling


Integrate & Manage
Automatically clean, integrate, and manage programmatic training data from all sources


Train & Deploy
Train and deploy state-of-the-art machine learning models in-platform or via Python SDK


Analyze & Monitor
Analyze and monitor model performance to rapidly identify and correct error modes in the data

Solutions —

AI Use Cases

Build and deploy use cases previously blocked by training data by combining state-of-the-art ML with industry-specific best practices and flexible API-based integrations using Snorkel Flow.

Risk Classification
Classify policy documents on the basis of the behavior or occupation to assess risk.
Customer Segmentation
Build customized promotions by analyzing customer behavior and demographics.
Clinical Trial Matching
Determine clinical trial candidates by categorizing patient records.
News Analytics
Extract entities, events, and relationships to improve investment and risk strategies and more.
Interaction Analytics
Understand every customer interaction deeply by analyzing chats, emails, and tickets.
Financial Spreading
Manage credit risk by collecting financial and non-financial data in any format from statements.
Search Engine Optimization
Identify named entities in customer search queries and optimize content on websites.
Product Recommendation
Enhance recommender systems by identifying entities (price, keywords, etc.) in product descriptions.
Contract Intelligence
Extract and organize data from a wide variety of complex contracts efficiently.
Product Catalogs
Extract product attributes from tables, lists, and forms for cataloging.
Email Filtering & Routing
Classify emails to remove spam and route queries to the correct channels.