Vanderbilt's Open Source Amplify GenAI Enterprise Platform

Summary

The video script details the development of an open, enterprise platform at Vanderbilt University for generative AI experimentation across disciplines. It emphasizes the importance of vendor independence, data control, and cost-effective scalability. The platform allows for model selection, quiz creation, template sharing, and integration with Amazon Web Services. It also showcases capabilities for diagramming, visualization, and assistant creation, all aimed at enhancing operational efficiencies and fostering innovation.

Takeaways

• 😀 The video discusses the implementation of an open, enterprise platform at Vanderbilt University to explore generative AI through a chat interface across various disciplines.
• 🔍 The platform is designed to be vendor-independent, allowing for the use of different models that are rapidly evolving in the field of generative AI.
• 💾 Vanderbilt has developed an open-source software system that can be downloaded, set up, and run within an organization's Amazon Web Services account.
• 📚 The platform includes a basic chat interface for generative AI, enabling users to select from different models and vendors.
• 📝 The script demonstrates creating a quiz on generative AI using the platform, showcasing the ability to customize and generate content in a specific format.
• 📑 The platform allows for the creation of templates, which can be used and shared across the institution to streamline content creation.
• 📈 Vanderbilt has deployed the platform at scale, emphasizing the importance of controlling data and integrating it easily.
• 💼 The platform addresses cost concerns by providing a cost-effective solution for institutions at scale.
• 🤖 The script introduces 'assistants', which are vendor-independent implementations similar to open AI GPT, with pre-set knowledge and guardrails.
• 📊 The platform supports various functionalities including quiz creation, template sharing, and the ability to generate diagrams and visualizations.
• 📋 The video also covers the creation of custom chatbots and the use of guardrails or custom instructions to control the responses of these bots.

Q & A

What was the main goal of creating an open Enterprise platform at Vanderbilt?

The main goal was to provide a platform where people from different disciplines could try out generative AI through a chat interface and innovate in various ways.

Why was it important for the interface to be independent of any particular vendor?

It was important to ensure flexibility and the ability to accommodate the rapidly evolving models from different disciplines without being tied to a specific vendor's offerings.

What were the key considerations for Vanderbilt when implementing the generative AI platform?

The key considerations included having control over their data, the ability to integrate data easily, and finding a cost-effective solution at scale for the institution.

Can the platform developed by Vanderbilt be used by other organizations?

Yes, the platform is open-source software that can be downloaded, set up, and run within an organization's Amazon Web Services account.

How does the basic interface of the Vanderbilt platform compare to other generative AI chat interfaces?

The basic interface of the Vanderbilt platform is similar to other generative AI chat interfaces, allowing users to select different models from various vendors.

What customization options are available for creating a quiz on the Vanderbilt platform?

Users can customize the quiz by selecting different models, defining the format, inserting questions, and choosing options, which can then be refined or used as a starting point.

How can the Vanderbilt platform help in sharing and reusing prompts across the campus?

By turning prompts into templates, users can easily share and reuse them across the campus, streamlining the process of creating quizzes or other content.

What is the purpose of creating assistants on the Vanderbilt platform?

Assistants, which are vendor-independent implementations similar to open AI GPT, are created to provide guardrails and knowledge to perform specific tasks, such as creating presentations or answering policy questions.

How does the Vanderbilt platform support the creation of diagrams and visualizations?

The platform has built-in capabilities that allow users to create diagrams and visualizations, enhancing the exploration of information both in text and visually.

What is an example of how the Vanderbilt platform can be used to analyze documents?

The platform can be used to upload and analyze contracts, identify unusual clauses, and even extract information to populate an email client for further action.

Outlines

00:00

🤖 Open-Source Generative AI Platform at Vanderbilt

The video script discusses the creation of an open, enterprise-grade platform at Vanderbilt University that allows various disciplines to experiment with generative AI through a chat interface. The platform is designed to be vendor-independent, ensuring control over data and cost-effective scalability. The platform is open-source, allowing organizations to download, set up, and run it within their Amazon Web Services account. The interface is similar to other generative AI chat interfaces, enabling users to select different models from various vendors. The script provides a demonstration of creating a quiz on generative AI using the platform, showcasing the ability to customize and generate questions, as well as the option to save prompts as templates for future use. It also touches on the platform's capability to export content to Word documents and create visual diagrams and visualizations.

05:02

📈 Building Vendor-Independent AI Assistants and Visualizations

This paragraph delves into the creation of AI assistants that are independent of any specific vendor, with examples of how Vanderbilt has implemented these assistants to aid in tasks such as creating PowerPoint presentations and answering travel-related questions. The script demonstrates how to use an assistant to generate a presentation outline on the generative AI roadmap at Vanderbilt, which can then be converted into a PowerPoint presentation using a pre-baked template. Additionally, it shows how to build an AI assistant that is tailored to the university's travel policy, enabling it to answer policy-related questions accurately by referencing specific document sections. The paragraph also mentions the ability to create custom instructions for AI assistants and the platform's support for various document types, including the generation of study guides and knowledge check questions.

10:04

🛠 Customizing AI Assistants with Document Uploads and Guard Rails

The script explains how to build custom AI assistants by uploading relevant documents and setting up guard rails or custom instructions. It illustrates the process of creating an assistant called 'myTravelHelper' based on the university's travel policy, which can answer questions about expenses by referencing the uploaded policy document. The paragraph also introduces the concept of custom instructions, which allow for the creation of thought-provoking questions instead of direct answers, thereby adding an educational or reflective dimension to the AI's responses. Furthermore, it discusses the platform's organizational features, such as the ability to create folders and workspaces to manage and share chats, assistants, templates, and other resources across the university.

15:04

📑 Leveraging AI for Contract Review and Information Extraction

In the final paragraph, the script highlights the use of AI for reviewing contracts and extracting unusual or important information. It provides an example of uploading a sample consulting contract and using the AI to identify any unusual clauses, such as a humorous stipulation about meatloaf and legal action. The AI's ability to quickly filter and extract relevant information from documents is emphasized, along with the option to convert the AI's responses into an email, demonstrating the platform's practical applications for operational efficiencies and exploratory avenues within an enterprise setting.

Keywords

💡Enterprise platform

An enterprise platform refers to a comprehensive and scalable system designed to support a wide range of business processes and functions across an organization. In the video, the speaker discusses creating an open enterprise platform for generative AI that allows people from different disciplines to innovate and experiment with AI through a chat interface. This platform is independent of any specific vendor, which is crucial for flexibility and adaptability in a rapidly evolving field like AI.

💡Generative AI

Generative AI is a subset of artificial intelligence that focuses on creating new content, such as text, images, or music, rather than just recognizing or analyzing existing content. The video's theme revolves around using generative AI through a chat interface to innovate and create quizzes, presentations, and other content. The speaker demonstrates how generative AI can be utilized to generate quizzes on topics like AI ethics and to create PowerPoint presentations.

💡Vendor independence

Vendor independence means not being reliant on a specific provider or service for a particular product or technology. In the context of the video, the speaker emphasizes the importance of having an interface that is independent of any particular vendor to ensure flexibility and the ability to use different models as they evolve. This allows the institution to adapt to changes and choose the best tools for their needs without being locked into a single provider.

💡Data control

Data control refers to the ability of an organization to manage and have authority over its data, including how it is used, stored, and integrated. The speaker mentions the need for data control to ensure that the institution can easily integrate its data with the AI platform. This is essential for maintaining privacy, security, and compliance with regulations while leveraging AI capabilities.

💡Cost-effectiveness

Cost-effectiveness is the ability to achieve the most significant outcome or value for the least cost. The video discusses the importance of finding cost-effective solutions for implementing AI at scale within an institution. This involves balancing the need for innovation and advanced technology with the financial constraints and resource limitations that educational institutions often face.

💡Open-Source Software System

Open-source software refers to software whose source code is available to the public, allowing anyone to view, use, modify, and distribute it. In the video, the speaker introduces Vanderbilt as an open-source software system that can be downloaded, set up, and run within an organization's Amazon Web Services account. This open-source approach promotes collaboration, customization, and cost savings.

💡Chat interface

A chat interface is a user interface that allows users to interact with a system or service through text-based messages. In the video, the chat interface is highlighted as the primary method for engaging with the generative AI platform. Users can select different models and generate content like quizzes and presentations through this interface, demonstrating the ease of use and accessibility of AI technology.

💡Template

A template is a pre-defined format or structure that can be used as a starting point for creating new documents or content. The speaker discusses creating templates for quizzes and other content, which can be customized and reused by different users within the institution. This allows for consistency, efficiency, and the ability to build upon previous work.

💡Customization

Customization refers to the process of adapting or tailoring a product or service to meet specific needs or preferences. In the context of the video, customization is shown through the ability to modify quiz formats, create templates, and build assistants that are specific to the institution's needs. This level of personalization allows for greater utility and relevance of the AI-generated content.

💡Visualization

Visualization involves the creation of visual representations of data or concepts to aid in understanding and communication. The video mentions the capability of the platform to create diagrams and visualizations, such as flowcharts for university admissions processes. This feature enhances the way information is presented and helps users to better grasp complex processes or data.

💡Assistant

In the context of the video, an assistant refers to a specialized AI model that is built with certain guardrails and knowledge to perform specific tasks, such as creating PowerPoint presentations or answering travel policy questions. The speaker demonstrates how these assistants can be created using uploaded documents and custom instructions, providing examples of how they can be used to automate and streamline various tasks within the institution.

Highlights

Vanderbilt University aimed to provide an open, enterprise platform for generative AI experimentation across various disciplines.

The platform needed to be independent of any specific vendor and support rapid model evolution.

Data control and easy integration were critical for the university's data management.

Cost-effective scalability was a priority for implementing AI at an institutional level.

Vanderbilt developed an open-source software system deployable within Amazon Web Services.

The interface allows users to select from various AI models for generative tasks.

An example of creating a quiz on generative AI using a specified format was demonstrated.

Customization and iteration of generated content, such as quizzes, are possible through the platform.

Templates can be created and shared for standardized tasks like quiz creation.

The platform supports exporting content to formats like Word documents for distribution.

Diagramming capabilities were integrated for visual representation of processes.

Visualization tools are available for data representation within the platform.

Pre-baked assistants were created for specific tasks, like creating PowerPoint presentations.

Assistants can be built with custom guard rails and knowledge, such as travel policy adherence.

Document upload and utilization for creating quizzes or study guides were showcased.

Custom chat bots can be rapidly developed using uploaded documents and policies.

The platform allows for the creation of custom instructions to guide AI responses.

Folders and workspaces can be organized for better management of AI interactions.

Control over AI model settings, including temperature and output length, is provided.

An example of identifying unusual clauses in a contract using the platform was given.

The platform's ability to filter and extract information from contracts was demonstrated.

Transcripts

one of the things we wanted to do on

campus at vanderbelt was provide a open

Enterprise platform so that people could

go and try out generative AI through a

chat interface across different

disciplines and go and try to innovate

in all different kinds of different ways

so one of the things that we realized

really quickly was that lots of

different disciplines wanted different

models and the models were rapidly

evolving so we wanted an interface that

was independent of any particular vendor

the other thing that we realized is we

wanted to make sure that we had control

of our data and the ability to go and

integrate our data easily and then the

other final thing that we really

realized was that cost was going to be

an issue and we wanted to find a way to

do it more cost effectively at scale for

our institution so vanderbilt. is the

platform that we put out it's also an

open- Source Software System that you

can go and download and set up and run

for your organization within your Amazon

web services account it's been deployed

at scale within

vanderbelt this is the basic interface

and it looks a lot like other chat

interfaces for generative AI you can go

and select from many different models

from many different vendors I'm just

going to start with open AI gp4 turbo

and I'm going to go in and I'm going to

say please create a

quiz for me on generative AI um please

produce your output in the following

format and then I'll just give it a

format for it

question

um insert

number and I'll tell it insert question

here

and we'll have several

options so I'm going to go and create a

quiz and get it to produce it in this

format now it'll go and run it will will

call out to the models in a private way

it'll start generating the quiz for us

using the format that we told it to now

there's lots of ways that we could go

and customize this quiz we could refine

it it's a great way to just generate

questions and you can look at them of

course and decide what you're actually

going to use and not use or use it as a

starting point for iteration it'll keep

going in generating the the questions

but I'm going to go ahead and cut it off

right here now let's say that in my

institution not everybody knows how to

go and format the quiz questions like

this and they want to be able to share

that so I'm going to go and copy this uh

prompt that I created and I'm going to

turn it into a template that we can use

by other people so I'm going to call it

quiz Creator um with format I'm going to

go and paste The Prompt in Here and Now

what I'm going to do is I'm going to

turn it into a template that others can

use and I'll just start off by putting a

placeholder so I'm going to put a

placeholder for the topic right here

that we can then go and and customize so

I'll save that and now we have this quiz

creator with format over here and if I

click on it it doesn't require me to go

through all the prompt engineering from

the beginning so now I might say like

generative AI ethics is what I want to

quiz on and maybe I want to switch the

model to um Claude 3 Hau and I could go

and do that here and now rerun The

Prompt different um different contacts

different topic that it's going to be on

not sure if we want the AI deciding the

ethics of the AI but we'll see what

happens

now we see it's already filled in the

prompt it's now going and using the same

template that we created before and I'm

going to just cut this off and stop it

now what's useful about this is as I

innovate in my domain and discipline I

can go and share this across campus for

example I could share share this with

Alan KS who I collaborate with and I

won't type out his old whole email

address but it's at vanderbilt.edu and

then I'm going to stop that and I could

share that with him and have it shared

or go any number of users and share it

with them so they can go and try out

that prompt the other thing I might want

to do is I might want to export this to

a Word document for example so I can

give this quiz because I think it's good

I want to you know actually give it out

maybe I want to go print it on paper and

so I can go and actually export it as a

Word document and I'll download that now

and then I'll upload um I'll open up

this word document now if you give me a

second and now we've got our um quiz

that's been exported I could also go and

you know do other things and one of the

things that we realized a lot of people

want to be able to go and do things like

create diagrams so we taught ours with

some prompt engineering and some

different capabilities that we created

to be able to go and draw diagrams which

can be a nice capability such as drawing

flow charts for different things like

you know University admissions processes

and so we have a number of different

diagramming capabilities built into this

so that we can go and explore both in

text but also visually what's happening

similarly we have the ability to do

visualizations which we see here are

some of the visualizations that we've

created um within this these are all

just sort of simple things that are now

available now one of the things we

wanted to do also was pre-bake a bunch

of assistants now what is an assistant

an assistant's kind of like an open AI

GPT but it's a vendor independent

implementation of the same sort of idea

where we put some guard rails and

knowledge into an assistant and then we

can go and do things I'll show you some

examples of the assistants that we've

built and some of the base ones that we

thought um we would create for people to

help them out so I'm going to do one

that's a popular win which is cre for

creating PowerPoint I'm going to create

a PowerPoint presentation on generative

AI road map at vanderbelt now it doesn't

know our road map but just for fun I'm

going to let it choose uh what our road

map should be or what the presentation's

going to be and I'm going to go ahead

and do this with Claude 3 opus

and now what we'll get is it will go and

start building out the presentation for

us in terms of an outline so what it's

doing is it's outlining the the

presentation in a format that we can

then convert with that assistant into a

PowerPoint presentation in our templates

and I'm just going to go ahead and cut

it off right here this is probably

enough for us to get a sample PowerPoint

presentation and I'll go and download

this except this time I'll convert it

into PowerPoint and use our vanderbelt

template that we've pre-baked into this

now we've got our download of our

presentation and we can go and up open

it up and now we've got generative AI

roadmap at vanderbelt built into our

PowerPoint um template and so we can

rapidly go and start creating content

that's an example of one type of simple

assistant another type of simple

assistant is one to go and think about

answering travel questions so for

example we can go and talk to this

vanderbelt Travel Pro and I'm going to

talk to it with gp4 and I'm going to say

please tell me the policy on first class

travel and now it's already an assistant

that it's already been built with our

travel policy and expense policy

document it's been told how to respond

and how to behave what it's doing right

now is it's using retrieval augmented

generation and our implementation of it

to go and search the policy find the

relevant sections quote them back to the

user like we see here and then provide a

potential interpretation of the policy

and then also who to go and follow up

with contact one of the other nice

things is we can go and look at the

actual excerpts from the policy that it

used for the basis of its answer exactly

which document it used but it's filling

out and helping people to now go and

answer questions first class seating is

reimbursable only if the Traveler's Vice

Chancellor and chief business officer

have approved the request in it writing

in advance of booking and that is on

page three of the policy and now we can

go and look at that check it out see

what we think and it lists different

interpretations now how did we go and

build something like this well it's

actually quite simple um first I'm going

to show you how we would upload

documents and use them and then I'll

show you how to come back and we'll

create an assistant so let's go back to

our quiz creation example where we

created this quiz originally I'm going

to copy that prompt and I'm going to

start over and pretend that I haven't

created that quiz and now what I'm going

to do is I'm going to upload a file and

I could go and choose any file on my

machine in this case I'm going to go and

choose an existing one that I've already

uploaded so I don't have to upload it

again and I'll just say I want to build

a quiz on the Travel policy I'm going to

take our travel and business expense

policy which has been uploaded already

and is now available and I'm going to

create a quiz based on this document and

policy now we don't see the document

like in the prompt itself it's separated

down here but um what we see is it's now

going and generating whoops I made a

mistake in my prompt I need to go and

tell it this shouldn't be on generative

AI ethics it should be a quiz for me on

the attached travel policy I need to qu

clarify it maybe a little bit confused

about what are we're supposed to do so

now we see who is the approval Authority

for the travel and business expense

policy you know for what reasons can

non-travel related business expenses be

in reimbursed first and so now it's

going and generating the quiz based on

the policy document that I uploaded we

can support PowerPoint word all kinds of

different policy document types but you

can also go and take a lecture you have

like a whole bunch of slides that you

might um you know be about to give in

class and you can say okay now generate

a study guide to go along with this as a

draft and then you as a faculty member

could take it and edit it or generate a

quick set of knowledge check questions

that I could go and look at and decide

if I wanted to use some of them or not

or guide it in the right direction or I

might ask as a student go and upload a

document ask it to challenge my

assumptions in my presentation and help

ask hard questions now I'm going to go

back to the assistant and I'm going to

build a new assistant we can do this

conversationally um or we can just do

the quick way I'll show you the Quick

Way first and then we'll go back and do

the same thing in a second here so we're

going to build an assistant called my

travel

helper um and we'll just make make it

really simple please answer questions

using the attached travel

policy and I'm going to go and select

one the

existing uh travel

policy and that's all it takes now I've

got an assistant that's based on my

policy I'm going to save it the

assistant will get created in a second

here which it's doing right

now and then I'll be able to go and chat

with it so now if we go over here we've

got my travel Helper and let's go and

talk with it um can I rent a um

scooter is that an invalid expense it's

already gotten our our travel Document

it's the my travel helper which that's

what tells us it says no you can't um

rent one on page five it states

motorcycle scitt rentals are not allowed

that's all it takes to grilled a custom

chat bot you can go give it as many

documents as want you want one or 30

build them all in and he can go and

search them and automatically begin

answering questions and you can go and

put guard rails in place for those

conversations we also support going and

creating your own guard rails which are

also called custom instructions

sometimes so we're going to create a set

of custom instructions and we'll call

them um never answer the

question um

um or actually and then we'll go and

we'll say no matter what never answer

the question instead pose a thought Pro

provoking question in response so this

is my guard rail that I'm put on I'm

going to call this a custom instructions

and save it and now I can go and I can

start a new chat and I'll go down here

to the never answer the question and

I'll say um what uh we'll say where is

vanderbelt University

located should know the answer to this

and it says what might be the reasons

University's location is significant to

prospective students and faculty and so

instead of going and answering my

question which is the default thing it

would do it's instead you know

specifying a followup thought-provoking

question theoretically and so we can go

and build guard rails the same thing is

happening in the assistant we can go and

preate create all kinds of guard rails

we have the ability to go and create

folders so you can go and organize

things so my chats um about travel we

could then go and move our different

conversations in here to keep everything

organized we also have the ability to go

and create workspaces so you can go and

create workspaces they have different

sets of folders and prompts and

everything you can share everything you

can share all of your chats you can

share all of your um different

assistants that you're creating you can

share all of your templates that you're

creating across um university you also

have full control over things like um

the model you're using and then if you

go and create a new one you can CH

create the temperature that it's using

you can also have some limited amount of

control over potentially output length

as well as search and many many other

things so I hope you'll um realize the

importance of having an Enterprise

platform for chat how it opens up all

kinds of new avenues for exploration and

all kinds of operational efficiencies

and as one last closing thing I'll give

you an example so I'm going to go and

I'm going to take a contract

so we'll have a sample Consulting

contract you can imagine you have all

kinds of contracts coming into a

university or another organization you

want to maybe have a first pass identif

it has anything weird that you

immediately need to know about or maybe

you've already reviewed it and you want

to say hey is there anything else that I

might have missed so we're just going to

say please identify anything unusual in

this

contract we'll upload the

contract and now it goes and identify

some unusual things in this 10 page or

so contract um and now we've of course

loaded this Contra contract with some

fun sentences in there that are buried

in different paragraphs so one of the

fun ones is that there's a clause in the

whole harmless section saying that if

the company uh serves meatloaf on

Tuesdays it forfeits any ability to

pursue legal action against the

consultant so just finding really

quickly acting as a filter and going and

extracting information all kinds of

useful things and then if I wanted to go

and you know turn this into an email I

could click here immediately populate my

email client with that response send it

off to somebody have you seen this crazy

thing anyways I hope you'll check it out

I hope you see the value of this for

your organ