Transcript
This transcript was autogenerated. To make changes, submit a PR.
This is the agenda.
We'll go into artificial intelligence introduction, AI revolution and
unstructured data, the tech stack for ai, chat bots, and the future of ai.
So there are a lot of artificial intelligence definitions out there,
but this is one of my favorites.
It says that artificial intelligence is the ability of a machine to
perform tasks commonly associated with intelligent beings, which means that
something that a human could do now, a machine can do because you have given
it that ability to dig in and find the right answers that it needs to.
This slide is very interesting.
It shows that artificial intelligence isn't something new, right?
It started when IPM developed this computer and, it fought against
Gary Sparrow in 1997 against him for this chess game, and this
computer played moves, which were so amazing and never seen by any.
Anyone or never played by anyone, and that's why Gary
kis Farro couldn't defeat it.
Then there were games which were developed and these games had its own way of
learning and fine tuning itself to be able to win all the points and and get
whatever it wants to win the whole game.
There were, social media and Netflix and all had recommendations
based on what you're watching, it was gonna recommend you.
Okay?
This is what you can watch next.
This is good for you.
This content is something you like or not like.
That's like artificial intelligence has been around us for a while and then came
chat, GPD and it just took everyone away.
Within five days, like you can see the graph over here, there are
these different applications and is showing like within how many days
did it hit that 1 million user mark?
And you can see chart.
GPT did it within five Ds, which is insane.
When it did that, before that Instagram did it in 2.5 months, the 1 million mark.
And then also there was thread which came in and within one hour
it was hitting 1 million users.
And within two hours I think it did hit 2 million.
But today we are not interested in threads is definitely an entrusting
applications for people who have used it.
But chat, GPT has its own different.
Use case, which is not just a social media tool like all these others,
most of it you can see are more for social media, entertainment purposes.
But Cha g PD is very much used by everyone.
It be a researcher, may be someone who's working in a corporate job
to make a jobs or lives easier.
You are going to use this application.
Next, we'll deep dig deeper into AI revolution and unstructured data.
Why did this whole AI revolution come into picture?
Because 90% of your data is unstructured, which means it is in the format of text,
images, audio, video, and you want to make sense out of this data because.
Data is important.
Any data you give someone, there is a way to make analysis, predict future, and come
up with some better solution that can.
That you cannot, where if you don't have that data, you can't analyze it.
So tradi traditional database struggle to process this unstructured data, and that's
where generative AI comes into picture.
It goes in, finds the whole scenario or what is happening in your data, gives you
the result and extracts value for you.
What is generative ai?
Generative AI is AI models that generate text images or insights.
The examples are GBD four LAMA Cloud.
Why is it powerful?
Because it's humanlike.
The responses and summarizations are amazing that you.
Even if a human had to dig into so much like unstructured data and
bring it, it's was gonna be harder.
So someone who can do this for you without any man, manual intervention is amazing.
Understanding your unstructured data.
There is a lot of unstructured data out there and all these
things that you see on the screen.
Videos, images audio texts fall under it.
The types are also varied, like email, chat logs clinical
notes, pathologic reports.
All of them fall down structured data.
And if you wanna make sense of it you can.
So why is it hard to process?
No predefined schema or structured.
That is why it is hard, because if there was a proper table and
you had like your columns into it, you could write a simple select.
S statement or write like a procedure and do the whole data science or
python analysis on it and predict the, what the data was bringing you.
But with this, it becomes difficult because there is no structure.
How Gen AI processes it is.
It has the natural language processing for text understanding.
It has summarization and document classification.
It has con conversational AI for chatbots feature, and then it
has rag and vectors and chalk.
So what this whole side means, it, the gen AI has those specific capabilities.
Like it has those necessary functions which can summarize your whole
essays, big, long paragraphs and pages of essays into something small.
It has, that conversational ability.
So if you ask it a question and ask it a counter question on the first
question, it remembers the first question and gives you the answer or
the context of the answer accordingly.
There is rag.
Rag is a very amazing feature wherein your data, unstructured data is
divided into chunks and it is given that specific, unique identifier for
it to be able to identify later on.
So overall, all these things make reduce the administrative burden,
enhance data driven decision making for you, and that's how generative
AI processes your unstructured data.
The role of gen AI is twofold.
The traditional one was very predefined, no learning and limited, whereas
AI power chat bot is context aware.
It is very much self-improving via, data and highly conversational.
The tech stack.
This is my short and suite, I would say like architecture diagram that you
can see wherein your whole flow of the.
Chat bot experience, you can see over your, so the first thing,
it starts with a question.
This question gets the embedding.
Embedding is a specific, unique identifier for your question based
on what question you're asking.
This embedding is then going and searching in your vector data.
And this vector data, again, is stored into embedding.
So like I was mentioning, your data, the rag.
Chunks your whatever files and all you have into specific embeddings
and stores it into vector database.
And whenever a match is found between this embedding and this em
embedding, that relevant chunk is returned and there is some engineering
done to make it look beautiful and then returned to the end user.
So that's the tech stack.
You need a large language model.
You need an embedding database.
There's some vector search functions you need to apply to be able to search.
And then the framework, which is more of the UI piece.
You can do it a stream data app or something.
So you can combine the both UI and the backend to it.
Here is an example of a chat bot that I built myself.
So what.
I did was I had this whole a whole stage wherein I had different PDFs in it.
One of them was my resume and I wanted to ask question on my resume.
So that's what I did.
And the chunk, I did the function for that I created like a whole vector
database, stored my chunks into it.
And also called my, for call this function to get the relevant chunk,
then did a getcha history function.
Summarize questions with history was used to remember the context
of the previous questions.
Complete function and prompt function was used to make the answer more better.
So these are the couple of function and the whole script and the UI
piece of it for the chat bot.
So building chatbot step by step.
Definitely.
We saw the whole flow and how chatbot can we build, but there is a little
bit of pre-processing and fine tuning piece to it as well, right?
Like you cannot just build it.
And that's about it.
The pre-processing comes with data ingestion.
So you definitely have to collect like all types of data from different sources,
bring it in a proper stage, location, do some data processing and transformation
on it because, so there is definitely gonna be some special characters which
are gonna throw your s Python script away.
So make sure you remove those special characters and then
build a model on top of it.
For that bot model.
You'll need like an LLM rack.
There'll be some fine tuning and testing you'll have to do because
once you've built it, the first.
Three to six months is training period, and there are questions being
asked by different users, is still trying to understand what is the
question and context and trying to make sense out of your data for you.
And then finally you'll deploy and monitor and give it to like the
specific end users that you want to.
The case studies for industry applications are unlimited.
There is no limit where you can use chat bots, right?
From customer support patient q and a, contract analysis, policy bots.
There, there are like finance bots available to, for fraud detection and all.
So any industry, if you have those unstructured data and you
want to make sense out of it.
It is possible with your chat bot.
Next steps to adopt chat gen ai.
I would say if you are trying to build an application, definitely start with.
Start slow.
Start with the crawl phase wherein you are bringing all your applications data,
you are storing it in a stage, trying to do the cleansing on it, running
some scripts to do the chunking and vectoring of your database and all that.
So that's like the first crawl phase for you.
Then comes the walk phase wherein you're building the whole decision
engine and you are deciding, okay, if, what are the key parameters which
you would want to track or which the users are gonna ask questions on.
How are you gonna understand okay, if this is answered, it is a good.
Answer versus if this is answer is the wrong, bad answer.
Those decision engines is the next phase.
And then finally, you are gonna deploy it and put it out there for everyone
to use and see how the changes happen.
I. Also there are a lot of features wherein the user can give feedback.
There are like plus thumbs up and thumbs down features that you see
in different chat bots out there.
You can incorporate those so that your model understand, okay, this was
a negative answer, and next time it, it makes sure what are the positive
answers and brings you answers based on that, rather than giving you the
negative ones or the wrong ones.
The future of AI Chat Botts we are definitely in the tip of AI chat
Botts, and there is so much more.
There are multimodal chat botts, which is voice and text related chatbots.
Combined, there is autonomous AI agents.
So here you can see there was we trip booking.
I asked.
The GPT four, it went in the high camp website, brought in the best deal for
me based on what dates I'd given it, and then told me this is like the best deal.
It has 95% rating.
Would you like to book it?
And if I say book it, it's gonna go ahead and book it for me.
So that's what an agent take.
AI would do it.
Will you just.
Have to type your questions in your chat words.
This is gonna go in the specific website, bring the answer for
you, and you are good to go.
Then there are AI powered workflows.
So a lot of manual processes are going to be replaced by ai.
So right from your data engineering step ingestion or the whole like data cloud
framework, building the whole, a raw gold and silver layer, the scripts will do it
for you without any manual intervention.
So those those like startup ideas are out there already and people
are working on it and building it.
And then there is gonna be a lot of AI governance and safety measures
that are gonna be built out there.
So even like companies who are very unsure about the usage of AI can securely.
Use AI under the whole governance and framework and stuff.
Next is we'll see what will the future be.
So I'm gonna leave you guys with this final slide, which is showing
two different perspectives.
One is a human from the wall e movie which had nothing to do and
grow weaker and weaker over time.
Versus on the other hand, there is this whole new phase
of discovery that we did with.
The AI revolution and we built amazing products and amazing solutions.
So it's in our hands.
Are we gonna be this side or that side?
And I feel like use AI and chat bots to your best use so that you can
do something which has not yet been discovered and discover amazing things.
Thank you.