© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.

Generative AI in seconds: Making

sense of unstructured data across
industries
Pawan K Mall
Senior Sales Engineer
Snowflake

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Popular Enterprise Generative AI Use Cases

Text analytics & Document

Conversational
generation chatbots
data analytics
Increase productivity of
Improve customer experiences Empower every user to
customer facing teams with
with accurate summaries and uncover insights without
chatbots for user guides,
analysis of reviews, service tickets, having to write code or
FAQs or wikis
chats, etc. analyze charts.
Democratize knowledge with
Reduce manual errors and employee Q&A chatbots that & more …
streamline processes with can quickly search multiple
automated document documents
processing

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Snowflake for Gen AI & LLMs: Platform Overview
USE AI IN SECONDS APPS IN MINUTES FULLY CUSTOM IN HOURS

Document AI Universal Search Snowflake Copilot Streamlit Custom UI Custom Orchestration

Use Complete, Embed, Vector Search, and more… Bring Your Own Make It Your Own Use Marketplace

Serverless AI & LLM Functions OSS LLMs Fine Tuning Partner LLMs
Amazon
Bedrock
Snowflake Cortex: Models & Search Snowpark Container Services

Snowflake-Governed Data

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Snowflake Cortex
A fully managed service that hosts and serves pre-loaded, on-demand AI models, LLMs and
vector functions accessible via SQL or Python - without having to manage the infrastructure.

Why Use It
Quickly and securely analyze your data and build AI applications Serverless Functions

contextualized with your enterprise data

translate() complete()
forecast() embed()
... ...

How To Use It Specialized General-Purpose Document AI Snowflake Copilot

Access the power of Snowflake Cortex via serverless SQL / Python

functions or as part of LLM-powered experiences such as Document
AI and Snowflake Copilot
Llama 2

Use Cases
• Cost effective way to build single tasks, including translating, com plete

summarizing and sentiment analysis

• Build more complex prompt engineering on foundational models
(i.e. Llama) in conjunction with vector search functions to build forecast

chatbot capabilities

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Snowflake Cortex: Key Takeaways

Easy To Use Flexible Cost Effective

Use AI in everyday analytics or Access industry-leading AI models, Compute optimized for inference
build LLM applications without LLMs and vector search and search to run where your data
infrastructure management functionality via SQL / Python is secure and governed
functions

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Snowflake Cortex: Specialized LLM Functions
Tuned to be more cost effective & performant for specific tasks. No prompt engineering required.

SUMMARIZE SENTIMENT EXTRACT_ANSWER TRANSLATE

Execute fast Easily detect Extract key Fast and

summarization customer information from continually
tasks that don’t sentiment (Positive, unstructured data improving
require Neutral, Negative) in a Question translations from
customization for customer Answer format any supported
analytics language

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Snowflake Cortex: COMPLETE Function
What Is It
LLM Inference Category Model Context Window
Serverless SQL / Python functions that run
inference on performant LLMs across every
model-size category COMPLETE( Large Mistral- 32k tokens
model: string, Large
messages: string
) Medium Mixtral-8x7b 32k tokens
Why Use It
Use LLMs from Mistral AI, Llama and Llama 70B 4k tokens
Snowflake Tables
Snowflake without moving data out of your
Snowflake account or having to manage ACCESS MODELS FROM Small Mistral-7B 32k tokens
infrastructure.
Gemma-7B 8k tokens

How To Use It
Use SQL / Python function in
Worksheets/Notebooks or Streamlit app
© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Process Unstructured Data with Document AI
Provides a first party multi-modal LLM through a visual Create an LLM
interface or SQL API to extract content easily from for Business
documents using natural language. Users:
● Ask questions in
natural language to
extract information
from documents,
Document: Scanned Images Of
Documents Documents
such as claims
reports, investor
DOCUMENT AI reports, and
prospectuses
Built-in, no-code
Documents With Handwritten application with a large
● Use UI to create fine-
Infographics Text language model
tuned LLMs specific
to their documents
© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Document AI: How It Works
DOCUMENT AI SIMULTANEOUSLY
COMPREHENDS ALL ASPECTS
OF DOCUMENTS:

SEMANTICS
Document type: Agreement of Sales

CONTEXT
Seller: Naomi Lynnea Burr

Buyer: Eric Drake, Tammy Drake

LAYOUT
Sales Price: 448,000.00

GRAPHICS USING MULTI-

MODAL LLM
Buyer’s Initials: Yes
Seller’s Initials: Yes

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Bring LLMs & GPUs to Your Data with Container
Services
Key Benefits
Open Source ● Bring the right LLMs to
LLMs your data that meets your
specific business use case
GPUs
needs

Container ● Host a range of services,

Services including GPUs for model
Third-party
Partner & Data training, full notebook
Inference
Proprietary LLMs Fine Tuning experiences (e.g. Jupyter,
SNOWFLAKE GOVERNED VSCode), and vector
DATA
databases (e.g. Pinecone,
ChromaDB)

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Build a front-end UI for LLM applications with
Streamlit
Key Benefits
● Turn data and models into
apps with open source UI
library for Python

● Deploy to Snowflake’s
secure and reliable
infrastructure in one click

● Maintain consistent
governance of data,
models and other objects
in Snowflake

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Demo

Call Centre Analytics with

Snowflake Cortex LLM and
Snowpark Container Services

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Call Centre Analytics with Snowflake Cortex LLM and Snowpark
Container Services
Key Outcomes

● Agents Efficiency KPI

● Call transcript
summarization

● Call Diarisation

● Call Intent

● First Call Resolution

Snowflake Data Cloud
● Customer sentiment
analysis

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved. Try out at: https://developers.snowflake.com/solution/call-center-analytics-with-snowflake-cortex-and-snowpark-container-services/
Thank you! Please complete the session
survey in the mobile app

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.
skillbuilder.aws

Build beyond
Redeem your free 7-day
trial of AWS Skill Builder

© 2024, Amazon Web Services, Inc. or its affiliates. All rights reserved.