Building a Banking Conversational AI
Key Points
- Conversational AIs use large datasets, machine‑learning models, and natural‑language processing to mimic human interaction, recognizing speech or text and translating intent across languages.
- Their core NLP pipeline consists of four steps: input generation (user voice or text), input analysis with NLU to determine intent, dialog management using NLG to craft responses, and reinforcement learning to improve over time.
- To build a banking‑focused chatbot, start by collecting existing FAQs (e.g., “Where do I find my account number?”), then map these to intents such as “access account” and train the model on varied phrasings.
- Define entities (e.g., account number, password, routing number) that populate the intents, and combine intents and entities into a dialog flow that delivers accurate, context‑aware answers to users.
Sections
- Conversational AI: Components Overview - The passage explains the four-step NLP pipeline of a conversational AI and outlines the initial steps for creating a banking‑focused chatbot.
- Building Conversational AI: Intents & Entities - The speaker outlines how to turn FAQs into intents, define related entities, and combine them into dialogs for use cases such as banking support, HR processes, and IoT voice assistants.
- Like, Subscribe, and Sandwich? - The speaker urges viewers to like and subscribe, then humorously asks if their sandwich has arrived.
Full Transcript
# Building a Banking Conversational AI **Source:** [https://www.youtube.com/watch?v=pOUBt-S5dHY](https://www.youtube.com/watch?v=pOUBt-S5dHY) **Duration:** 00:06:24 ## Summary - Conversational AIs use large datasets, machine‑learning models, and natural‑language processing to mimic human interaction, recognizing speech or text and translating intent across languages. - Their core NLP pipeline consists of four steps: input generation (user voice or text), input analysis with NLU to determine intent, dialog management using NLG to craft responses, and reinforcement learning to improve over time. - To build a banking‑focused chatbot, start by collecting existing FAQs (e.g., “Where do I find my account number?”), then map these to intents such as “access account” and train the model on varied phrasings. - Define entities (e.g., account number, password, routing number) that populate the intents, and combine intents and entities into a dialog flow that delivers accurate, context‑aware answers to users. ## Sections - [00:00:00](https://www.youtube.com/watch?v=pOUBt-S5dHY&t=0s) **Conversational AI: Components Overview** - The passage explains the four-step NLP pipeline of a conversational AI and outlines the initial steps for creating a banking‑focused chatbot. - [00:03:09](https://www.youtube.com/watch?v=pOUBt-S5dHY&t=189s) **Building Conversational AI: Intents & Entities** - The speaker outlines how to turn FAQs into intents, define related entities, and combine them into dialogs for use cases such as banking support, HR processes, and IoT voice assistants. - [00:06:15](https://www.youtube.com/watch?v=pOUBt-S5dHY&t=375s) **Like, Subscribe, and Sandwich?** - The speaker urges viewers to like and subscribe, then humorously asks if their sandwich has arrived. ## Full Transcript
This is a conversational AI in the form of a chat bot.
It's a sad and hungry story, but it does have a happy ending.
Now a conversational
AI
is something that uses a large volume of data,
a large volume of machine learning and a lot of natural language processing
with the aims to help imitate human interactions.
Conversational AIs can recognize speech and text inputs and translate their meanings across various languages.
They generate responses based on user intent.
So how do they work?
Well, there are four basic steps to the natural language processing that occurs within a conversational AI.
Now, the first of those is input generation.
This is where we, and by "we" I mean, like perhaps a hungry user awaiting a sandwich delivery -
We provide input in the form of voice or text through a website or an app.
Next is input analysis.
And this is used to decipher the meaning of the input and derive its intent
through NLU or natural language understanding.
Dialog management
is used to formulate a response in a way that mimics human speech using NLG.
Or natural language generation. And reinforcement learning,
well, that is used to refine those responses over time
based on the analysis of how well the conversational AI did this go around.
Simple enough.
So let's build one.
We're going to create a conversational AI that can work with banking queries.
So step one is to figure out the FAQs - or frequently asked questions.
Now, these are the FAQs from our end users, and chances are that they already exist someplace to like assist human customer service representatives, for example.
Well, maybe the reason listed on your website. Now common FAQs might be things like, "Where do I find my account number?"
You know, just sort of general questions like that.
Another one might be, "How do I activate my debit card?"
Things like that.
There's a large corpus of potential FAQs, but will start out just with a small segment of questions to prototype the development of this process.
Now, step two is to use those FAQs to form what's called intents.
So one intent might be - how to access my account.
From here, we'll teach the conversational AI the ways that a user may phrase or ask for this type of information.
So, "I forgot my password",
"How do I log in", and, "How do I sign up for online access?" --
they're all phrases related to this intent.
With intents defined, step three is to build out entities.
And these surround specific user intent.
So, for example, we can create an entity that say, we'll call it account information.
And list the nouns related to this entity, so that might be account number, password, routing, number, username, that sort of thing.
And then in step four, we put these elements together to create a dialog.
And this is the dialog with our end users.
The intents allow a machine to decipher what the user is asking for,
and then the entities are used to act as a way to provide relevant responses.
Now are banking conversational API is an example of an online customer support use case,
but there are many other use cases for conversational AIs.
So, for example, we could have things like HR processes.
And we can use a conversational AI with an HR process to optimize things like employee training,
onboard processing, and updating employee information.
IoT devices can also use conversational AI
through voice-based digital assistants.
And we can do things like auto-complete on search fields to really start you
typing a query and then have the conversational AI completed.
Look today, most AI chat bots and apps have a somewhat rudimentary problem-solving skill.
An end user is unlikely to be fooled into thinking that this conversational AI is actually a real human.
But they can help reduce time and improve cost efficiency on repetitive customer support interactions,
freeing up those human resources to focus on more involved customer interactions.
Whether it's resetting my bank account password, guiding me through an HR
onboarding process, or even figuring out what happened to my sandwich,
conversational AI is here to help.
And if you'd like to learn more, check out these related videos.
And if you enjoyed this video, please consider hitting the like button and subscribing to the channel.
Did my sandwich show up yet?