Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

1

The Data-First
Platform for
Enterprise AI
2

Executive Summary: Snorkel AI


MOTIVATION & TECHNOLOGY PRODUCT TEAM

• Hand-labeling training data is the largest • Snorkel Flow: An end-to-end ML • Team of x from leading tech companies,
bottleneck in AI/ML today development platform centered including 8 PhDs + faculty
• A new programmatic approach to labeling & around the part that actually

building training data developed over 4+ years matters–the data–based on this new

at the Stanford AI lab programmatic approach

TRACTION FINANCING ROADMAP


3

A 30,000-foot View of
the AI Market
GLOBAL 2019 SPEND
AI is the largest growth area for enterprise
IT spend1
$37.5B
Enterprises’ spend is focused on AI “I don’t need a GPT-3 art project or a
generic sentiment analysis model”
solutions custom-trained on their data,
for their problems & objectives
INFRA MODELS DATA

Two of the three building blocks to enable


this have been commoditized

The third hasn’t changed in decades–and is the new bottleneck for AI


4

AI Today is Blocked by
Training Data

TRAINING DATA

Building AI applications today requires armies of human labelers

Non-starter for private, high-expertise, rapidly changing real-world settings


5

Billions of dollars
of enterprise AI
investment are
blocked on the
training data
6

A New Approach:
Programmatic Labeling

TRAINING DATA

Key Idea: Programmatically label, build, and manage training data


7

The Snorkel Project


USERS AND SPONSORS

4+ year research project at the Stanford AI Lab resulting in 30+


publications and many production deployments
8

Snorkel defines a
fundamentally new
way to build AI
applications
9

Snorkel Flow:
A Radical New Way to Build AI Applications

LABEL & BUILD INTEGRATE & MANAGE TRAIN & DEPLOY ANALYZE & MONITOR
1 2 3 4
Data Programmatically Version and Serve data Custom ML Models to Close the Loop

Snorkel Flow enables a faster, more practical, adaptive ML development process


10

AI Can Revolutionize Software


if AI is Revolutionized First

CONVENTIONAL AI SNORKEL FLOW

BUILD AN Weeks to months of manual Hours of push-button


AI APP labeling by experts development

CHANGE THE
AI APP
Re-labeling from scratch Minutes of push-button editing

BUILD THE Start from existing labeling


NEXT AI APP
Start from scratch
resources

Snorkel Flow is a zero-to-one shift for most enterprise AI use cases


11

Snorkel Flow Unlocks a Whole New


Space of Application Building
Software 2.0 Applications -
Unlabeled Data

Software 2.0 Applications -


Labeled Data

Software 1.0 Applications


Competitor 3
Competitor 1

Competitor 2
Competitor 4

Snorkel Flow is the first platform for building Software 2.0


applications - without labeled data
12

The Future of Enterprise AI Is Collaborative

Business Managers

Subject Matter Expert Developer

Core Driver: Data Scientist

Snorkel Flow application templates enable scalable GTM motion & full
customer configurability
13

Snorkel Flow Applications Templates

Customer AI Customer AI Customer AI Customer AI


AI APPS (CUSTOMER BUILT) APPLICATION LAYER
Application Application Application Application

AI APP TEMPLATES
Snorkel Flow Application Template Snorkel Flow Application Template TEMPLATE LAYER
(SNORKEL/CUSTOMER BUILT)

SNORKEL FLOW PLATFORM PLATFORM LAYER

AWS Azure GCP On-Prem INFRA LAYER

Snorkel Flow application templates enable scalable GTM motion &


full customer configurability
14

An Empirically-Proven Way To
Accelerate AI

Customer 1 Customer 2

Social Media Medical Image Content


Use Case Use case Use case
Monitoring Labeling Classification

Human Effort Saved / Application


6 8 - 14 10 - 100K+ 1-3 3-6
(person months) Hand labels

Snorkel Flow has been proven to save person months-years (at or


above quality parity) on diverse applications

[Bringer et. al., SIGMOD DEEM ‘19; Dunnmon & Ratner, et. al., Cell Patterns ‘20; Bach et. al., SIGMOD Industry ‘19]

You might also like