# Beyond 101: Catalyst | Part 1 – Upgrade Retail Experiences using Catalyst Serverless + AI

## Метаданные

- **Канал:** Zoho
- **YouTube:** https://www.youtube.com/watch?v=7C88zn1JRi4
- **Дата:** 25.03.2026
- **Длительность:** 46:18
- **Просмотры:** 222

## Описание

In this session from the Beyond 101: Catalyst series, we walk you through how to combine Zoho Catalyst's serverless functions, QuickML LLM Serving, and AI capabilities to build smarter, more responsive retail applications — without worrying about infrastructure.

What You'll Learn:

* How to use Catalyst Serverless Functions (Advanced I/O & Job Functions) for backend logic
* Deploying front-end apps using Catalyst Slate — including GitHub auto-deploy
* Using Catalyst QuickML LLM Serving to integrate generative AI into your apps
* Scheduling automated tasks with Catalyst Cron Jobs (Job Scheduling)
* Storing and querying data using Catalyst Cloud Scale (Data Store)

Real-World Use Cases Covered:

1. AI-Powered Review Summarization — Automatically summarize customer product reviews using an LLM and a Cron job that updates in real time

2. Sales Analytics & Marketing Suggestions — Feed sales report data to an LLM to generate actionable marketing campaign recommendations

3. SEO Alt-Text Generator — Use Catalyst's Vision Language Model to auto-generate SEO-optimized alt text for product images

📌 This is part 1 of the Beyond 101: Catalyst series, which builds on the foundation of Catalyst 101. If you're new to Catalyst, we recommend getting up to speed with the Catalyst 101 playlist first.

CHAPTERS:

00:10 – Introduction
00:44 – Power of GenAI
01:17 – Harnessing LLMs for retail
03:54 – What is Contextual Intelligence (CI)?
04:31 – What is Serverless Computing?
05:11 – Catalyst platform features
07:08 – Demo: Review Aggregator App
09:08 – Slate – Front-end deployment service
11:04 – QuickML
12:32 – Job Scheduling
13:42 – Setting up the project
30:00 – Demo: Sales Analytics & Marketing Suggestions
32:47 – Demo: SEO Alt-Text Generator
37:16 – Zia – Context Intelligence Engine
40:32 – GenAI Actions
41:28 – Sentiment Analysis
43:52 – Smart Recommendations

Resources:

Catalyst 101 Playlist: https://www.youtube.com/playlist?list=PLlC7sQNISSUQqYzUz2flSG5niU_ylRw75

Catalyst Help Docs: https://catalyst.zoho.com/help

For more such developer-centric events, visit:
https://community.zoho.com/developer/events

If you're not part of the Zoho Community, and would like to network with other Zoho developers across the globe and benefit from all that the community has to offer, visit our platform given below. Sign in using your existing Zoho credentials or create a new account through the Sign Up form. 

https://www.zohocommunity.com/zcs/group/developers-zoho-user-group/

Drop an email to developer(hyphen)community(at)zohocorp(dot)com for further information.


👇 SUBSCRIBE👇

️ ♦️ YouTube - https://www.youtube.com/zoho
️ ♦️ LinkedIn -  https://www.linkedin.com/company/zoho-community/
️ ♦️ Twitter - https://x.com/ZohoCommunity
️ ♦️ Facebook - https://www.facebook.com/ZohoCommunity

💬 Have questions or want to suggest a topic for the next Beyond 101 session? 

Drop them in the comments or post in the Zoho Developer Community!

## Содержание

### [0:10](https://www.youtube.com/watch?v=7C88zn1JRi4&t=10s) Introduction

— Welcome to this session on upgrading your retail experiences with serverless and AI. So, we are very clearly in the middle of the generative AI era, right? Terms like generative AI, LLMs, transformers, context, and agents are not just buzzwords anymore. Uh they're being actively used and implemented in many apps today. And any application without these capabilities is at the risk of being considered out of date or left behind.

### [0:44](https://www.youtube.com/watch?v=7C88zn1JRi4&t=44s) Power of GenAI

And Gen AI brings immense power. Basically, it understands and generates natural language. It's incredibly versatile and more importantly, it continuously learns and improves. So, this means that the apps we build can get smarter every day. The Gen AI tech that we have today is already great, but what's even more amazing is that it is the worst it can ever be and it only goes up from here and it comes to capabilities. So, when harnessing LLMs for retail or

### [1:17](https://www.youtube.com/watch?v=7C88zn1JRi4&t=77s) Harnessing LLMs for retail

any other solutions that you want to build, it is critical to have a few things in mind for the implementation to be actually useful. The first is the business context. Your AI is only as good as your data, right? While they are adept at processing language, LLMs sometime sometimes often lack the specific understanding needed to fully grasp an organization's unique processes or industry-specific terminology or nuanced business objectives. And without this context, their responses may be generic or misguided, requiring human intervention to refine or validate the output for more business-critical applications. So, this critical capability of having a business context is an absolute must for us. And we all take data privacy and security very seriously. And since these models often require large amounts of data to function effectively, there is a risk of sensitive proprietary information being you know, inadvertently exposed during processing. And furthermore, unless properly managed, user data could be used for retraining or fine-tuning models, potentially leading to privacy violations. So, if you're going to leverage any Gen AI models, it has to be compliant with your strict privacy and data security requirements. So, always keep that in mind when deciding to use a particular LLM provider. The cost of providing this capability is also very critical. Just because we spent a lot of money on implementing LLM capability in applications or solutions does not mean that we could pass that cost on to our customers. So, whatever solution that we build must remain capable and useful while also ensuring that the affordability of our apps do not change for our customers. And then finally, it is very important to choose the right balance between capability and compute complexity or cost. So, for most use cases, you probably do not need a model with a very large context window or something that is very expensive to run. So, depending on your needs, it is important to choose something that is as lightweight as it can be. To have the right kind of LLM-based solutions, it's not just AI that you

### [3:54](https://www.youtube.com/watch?v=7C88zn1JRi4&t=234s) What is Contextual Intelligence (CI)?

want. What you need is CI and CI is nothing but contextual intelligence, which is AI combined with your business context. And today we will see how you can build these contextually intelligent solutions using Catalyst, which is a full-stack serverless development platform from Zoho. Now, for those who attended the previous sessions on Catalyst or someone who has explored Catalyst would already know what this means, but for the freshers here, what exactly do I

### [4:31](https://www.youtube.com/watch?v=7C88zn1JRi4&t=271s) What is Serverless Computing?

mean by serverless computing? Serverless computing basically enables developers to focus entirely on business logic rather than worrying about server management, provisioning, or scaling. Basically, Cat- Catalyst handles the entire infrastructure automatically for you, which translates to no more manual setups, no managing of virtual machines, or no downtime during traffic spikes. So, for your business solutions, this translates directly to faster development cycles, lower operational overhead, and a cost-effective paper use model aligning directly with the actual usage.

### [5:11](https://www.youtube.com/watch?v=7C88zn1JRi4&t=311s) Catalyst platform features

When it comes to the Catalyst platform itself, these are all the capabilities that we provide out of the box. So, in the compute space, we have functions and app say. And we have com- components and services that handle web and mobile deployment as well. And on the DevOps side, we have our own CI/CD capability so that you can have your own pipelines running for your applications. When it comes to integrations, we have our event bus service called as signals. And then we have our own AI/ML services that we will explore shortly. And then on the back-end side, for example, we have our own back-end-as-a-service offerings, which include data store, Stratus, which is our own object storage solution, and so on. Of course, on the security side, we also have API gateway and authentication and all these components. — [snorts] — The key thing about Catalyst is that it works with any tech stack that you use. These are just some of the examples that I have you know, displayed here. But you know, whether you're using Next. js, Angular, React, or on the back-end side, you may be using Java Spring Boot or Django or you know, Node Express, whatever framework that you use in your day-to-day development, Catalyst supports those frameworks out of the box. — [snorts] — Now, to better understand how you can implement LLM capabilities in your retail solutions, I have built a few demos in Catalyst and let's take a look at them. So, what I'll do is I will build one application from the scratch and then we will take a look at a bunch of pre-built demos that I have already finished building and deployed on Catalyst.

### [7:08](https://www.youtube.com/watch?v=7C88zn1JRi4&t=428s) Demo: Review Aggregator App

The first one that I would like to showcase is a review aggregator app. Imagine someone coming to your e-commerce store looking to buy a product. There will be reviews from many people and it's extremely useful for your customers to make a buying decision if there is a summary of what all the reviews are talking about. LLMs are extremely good at this and that is what we are going to do. leverage and use in this demo. So, before I move on to the Catalyst console, maybe I will show you what the app is going to look like. — [snorts] — Okay. So, this is the application and we have a sample product page here for the PlayStation 5. And we basically have a list of users who have uploaded their reviews for this product. Some are positive, some are negative, some are somewhere in between. And here you can see that there is a summarized review. So, this is something you've seen in Amazon or other stores as well. So, the AI basically summarizes the entire list of reviews that customers have shared so that people who want to buy this product can you know, just go through this and make a decision on what they want to do. So, this is exactly what we will be trying to build. So, this is like a very small proof of concept, right? Now, I am going to switch to the Catalyst console. So, this is the index page of the Catalyst console. I'm going to go ahead and create a new project. I'm going to call this the 101 aggregator because this is a review aggregator. — [snorts] — And I have created the new project. Got it.

### [9:08](https://www.youtube.com/watch?v=7C88zn1JRi4&t=548s) Slate – Front-end deployment service

All right. Now, for this particular session, we will be exploring a few new services in Catalyst. The first one is Slate. Slate is basically our front-end deployment service. If you have a front-end that is built in whatever framework that you have, Next. js, Angular, Ember, React, or even just static websites, you can use Slate to build and deploy it here. And once you've deployed it, you basically get a URL that you can use to access the application. So, there are many ways in which you can deploy your application. One is obviously from the CLI. If you have your Catalyst front-end project set up using the Catalyst CLI, you can deploy it through there. Or you can upload your build zip file directly. Or you can also connect your GitHub like I have and pick a particular front-end repository. And add a deployment name, select the branch that you want the service to listen to, and then if you want, you can enable auto deploy. So when you enable auto deploy, what happens is that whenever a commit is made into the branch that you select here, Catalyst will automatically detect that, run the entire build process for your front-end app, deploy it, and then give you a URL. So this is a really powerful service that we launched recently and it makes it really easy and convenient for you to you know manage all your front-end applications. Of course, for this demo, I will be deploying it through the CLI, which we will take a look at in a while. And the next service that we will be using in this particular demo is Catalyst

### [11:04](https://www.youtube.com/watch?v=7C88zn1JRi4&t=664s) QuickML

QuickML. So Catalyst QuickML is a service that contains a lot of Catalyst core AI capabilities and our focus is specifically going to be on LLM serving. So Catalyst has LLM serving capabilities built in, which means if you want to integrate any kind of LLM or generative AI inside the solutions that you're trying to build, you don't have to connect a third-party provider who provides these LLM capabilities to you. So Catalyst provides this out of the box and we have four models here that you can pick from and pick from based on your need and requirements. So we have a model that specializes in you know text, image, and video understanding. So one is like a high-performance mixture of experts model, one is good at coding, and the other one is basically an instruction-based model. So depending on your need, you can pick one and use it. Say for example, if you want to use the Gwen 14 billion model, you click on it and it is going to show you how the API details is going to look like, the OAuth scoping, and then a sample showing you how you can implement it in your code. You also have a sample request and response here as well for your reference.

### [12:32](https://www.youtube.com/watch?v=7C88zn1JRi4&t=752s) Job Scheduling

reference. And the third service that we will be using is our job scheduling service. So the job scheduling service, we are going to use it here primarily to run a cron job that will run at a regular interval. The purpose of having this cron job is to run it say every 5 minutes or every 15 minutes or 20 minutes to read all the reviews that the customers have posted and then send that review data to the LLM, which will then generate and summarize the review. So the reason why we have this as a cron job is because customers keep adding new reviews, right? And you want the summary to be updated over time. So that is the purpose of having this as a cron job and you can schedule a job function to run at a specified interval. So that is what we will be doing. — [snorts] — All right now. So before I come and set up the database tables and then the cron jobs, let us go and set this project up in my local machine as well.

### [13:42](https://www.youtube.com/watch?v=7C88zn1JRi4&t=822s) Setting up the project

well. So I'm going to switch to my terminal. And I'm going to create a new project. So for that, I'll just create a new empty folder. Catalyst 101 aggregator — [snorts] — And let me check into that. All right. I will now initialize the Catalyst project. And we will pick the project that we just created. So we will be using functions and slate for this project. So we will be needing an advanced IO function and also a job function. So first I will select advanced IO function and then I will add the job function next. I will select Node. js 20 as the runtime because that is the programming language that I'm going to use, but you are welcome to use anything that works for you. So I will call this function the review push function. We'll leave the rest at defaults. And then it is going to ask you to select a framework for your front-end. So to keep it simple, I'm just going to use static, but of course, you can use anything that you want. You can give a name for your front-end app. I'm going to call this aggregator. We'll give the deployment name as default and the initialization process is complete. Now like I said, we will need two functions. So let me go ahead and add the second function as well. Catalyst functions add and this time it will be a job function. — [snorts] — I'll call this review aggre- aggregator function because this is the one that is going to summarize all the reviews. And we'll install all the dependencies. And we are good to go. Now let me switch to Visual Studio Code and open the project. All right. Now the Catalyst init process has set up both the uh functions and also the client for us. So the client as you can see, we just have some boilerplate HTML, CSS, and JavaScript. And we have something similar set up for the job function and also the advanced IO function here as well. Now I am not going to write the entire code from scratch because that is not the focus of uh you know this session. The session's idea is to explore the Catalyst services. So to make it easier and faster, I will copy-paste the code that I have already written here and I'll just do a quick walk-through. So I'll start with the index. html. I'll replace this. And uh I will also paste the main. js. So nothing really fancy going on here. I'm sure you know you all will figure this out when you try to build solutions yourself. So some basic styling and JavaScript going on. And you can see that I have basically defined the API call here on the front-end side to push the reviews. So basically every time a customer adds a review in the application, we go and store that in the database, right? So for that, I've written the API handling. And now let me quickly do the same for the advanced IO and the job function as well. So first I'll start with the advanced IO function. So I'll copy-pasted that and I basically defined two endpoints here in the back-end. So this is our back-end serverless function. So the post API is for storing the actual review data in the table. So we will have to create a table called as reviews in the Catalyst console and then store the required information. So basically I'm using the Catalyst SDK to insert the data inside the table that we will create. And then we also have a get method here basically to fetch all the reviews and then display it. So that when the front-end makes this API call, the list of reviews will be displayed in the application. Now I have extensively covered data store and the Zoho Catalyst query language in the 101 series. So I highly recommend that you take a look at it if you are not familiar with that. Next step is getting the job function ready as well. So let me go ahead and replace that. All right. So I have pasted this and as you can see, I'm making an API call to the Catalyst LLM service. And I've also constructed a prompt here basically telling the model its instructions on how to generate the review summary. And yeah, the rest is basically me um yeah, you know, inserting the records into the summary table. So, this is the front end and the back end configured uh code-wise. Now, let me just go ahead and uh quickly uh deploy it. Of course, you can also test it from your local machine using the Catalyst uh serve command. Uh but you know, let's just go ahead and uh deploy it. So, Catalyst deploy. So, it's going to go ahead and deploy both the functions and the front end. And we'll give it maybe a minute to do it. All right, it seems to have uh finished the process. All right, we also have the URL for the review push function, basically the function that is going to push the data into the review data into the reviews table. So, let me just change this endpoint in my um front end code because the code that I copy-pasted uh contains the details of the other project, which I will just quickly replace here. So, that uh you know, it works. All right. Uh let me just do the deployment again. Okay. So, it also says that the front end deployment process is also live. So, let's quickly go to the Slate console and then check. And you can see that it is now processing our front end code that we just deployed. And clicking on it shows that the status is a success. And we also now have a URL for the front end. So, let me close the old app that I had open. I will copy this URL and [snorts] uh paste it. And you can see that the uh application is now live. Of course, you can see we have something that says error loading data. That is for uh two reasons. One is we have not yet uh configured the tables in the Catalyst console yet. So, the API call that it is trying to um make uh will not work. And the second is that we also need to enable the CORS policy in the Catalyst console so that our serverless back end can also can be accessed from the specific domain of the uh front end that we have used here. So, let's do uh both of those real quick. Let me go to the Catalyst uh console and switch to Cloud Scale. — [snorts] — Cloud Scale is where our uh data store service exists. And we will basically create two tables. So, the first table will be called uh reviews. — And we will have a set of columns in this. The first column will be the product ID. So, this is basically a string. Uh you know, may or may not use this. We can skip this for this particular demo as well because we are not having multiple uh products here. But if you were to scale it, you'll either have product ID or name as the unique identifier for uh the products that exist. So, for us it in this case, it will just be the PS5. But I'll still add it here anyway. And I'm going to create a new column called as uh review. Sorry, reviewer. Uh that's going to store the name of the person who is adding the review. And uh I will use uh another column called as uh comment. And that will also be text. So, all of these uh for all of these columns, the data type will be text. And uh finally, we have uh something called as rating. So, this will be a number. So, I'm going to pick integer as the data type. All right, this column is done. Let us go ahead and create a new table for storing our summaries. — [snorts] — And this one uh will again have uh two columns. Which is product ID in case you have multiple products. And summary to store the summary text. Okay. So, with that, our table configuration is done. The next step is, like I said, we need to uh allow the front end domain to access all the uh back end functions that we have configured. So, for that, let me go to Cloud Scale again. Click on authentication. So, just go ahead and click on set up. We'll skip through this. And this is the part that we are focusing on, which is uh the authorized domain. So, you click on add domain. Enable CORS and then uh copy-paste the uh domain name of your front end. So, in my case, it will be this. So, I'll just go ahead and configure this. And you can see that it says CORS enabled. And uh finished. So, what I will do now is I will set up the job scheduling function as well. And uh let me go ahead and create a job pool. So, I'll call this the uh aggregator job pool. And I'll set up the memory for this. And I'll create a cron function. I'll call this the aggregator cron. And we'll make this a recursive function that runs uh every uh say 2 minutes. So, of course, the frequency is up to you to decide. And I'm going to call this aggregator job. Aggregator job pool. I'm going to pick this function. And uh click on create. Okay. Uh let me just go ahead and uh quickly ensure that my API endpoints are correct for the LLM. Okay, let me uh change the URL path here as well. And then uh deploy again. Okay. So, let me add in a few reviews here. Great console. I like it. And you can see that the data is uh available. And I'll maybe give a negative review for packaging. And once this is done we will now go and check our uh jobs here. And you can see that the uh job process has finished running. All right. So, now you can see that it has updated with the negative review as well saying, "While the console is celebrated for its strong exclusive, a few customers, like one reviewer, have complained about the poor packaging quality. " So, basically, we have used the cron function. And uh you can see that it has executed twice now. So, we asked it to uh run once in every 2 minutes, and it has run twice. And then after uh 2 minutes, it will run again and will keep updating the reviews. So, yeah, this is one way where you can combine different Catalyst services along with its uh LLM capability to solve different use cases that you may have uh in uh retail. Like I was saying, uh this is one particular use case that you can uh solve uh using the different Catalyst uh services like uh Slate, uh LLM, Cloud Scale, and uh job scheduling. I also have the demo for uh a few more use cases as well, but I won't be building them from the scratch. I will just show you the demo to give you an idea of how you can combine these different services together. The next one is of course analytics. So

### [30:00](https://www.youtube.com/watch?v=7C88zn1JRi4&t=1800s) Demo: Sales Analytics & Marketing Suggestions

you can also use LLMs to draw insights from reports like things that we may miss at a first glance. Say for example, let me open a new tab and open this application that I have pre-built. And this is a mock report that I have built containing the sales data for a company that sells electronics basically. Now you can feed all of this data to an LLM and ask it to generate suggestions for running marketing campaigns which you can use to you know improve sales in the underperforming areas. Say for example, if I click on analyze and suggest, let me just open the inspect tab to uh show you the API call being made. I'm clicking on analyze and suggest. So you can see that it has made the call and it has sent all the relevant data like category-wise performance, the regional split and the revenue from that region. And you can see that the LLM has given us a response as well. That you can see here in the pop-up model. So it basically gives you three marketing campaigns that you can run. Like introducing a bundle deal that you know includes a variety of smart home devices, the target audience for this and also you know the impact that you can expect like increase in revenue in the next quarter or something like that. So as far as the project is concerned, in the back end it is using the same components that I showed you. So it is using Catalyst Slate for the front end and a serverless function for basically processing the data that you send. And then I'm using the Catalyst LLM API to send all of the data to the AI model and it will just return that response back and we will show it back. So of course I have specifically um you know, given the AI model the function that I'm using in the front end to pass the data so that it sends it in a specific format. So you can see that I have also added it as part of the prompt here so that it does not send it in a way that is incompatible with how my front end code is expecting it. So this is one use case and the other

### [32:47](https://www.youtube.com/watch?v=7C88zn1JRi4&t=1967s) Demo: SEO Alt-Text Generator

use case that I would like to show you all is SEO optimization. So this is pretty simple and straightforward. So when you have an e-commerce store or if you're selling products in marketplaces, you will upload product images and for all those images you will have to generate alt text, right? And that alt text is important both for SEO purposes and also for accessibility reasons. So what I've built here is a quick little alt text generator. Now I have used a separate front end and then processing it by making an API call to the back end and then displaying it here, but in reality you can wire this up in such a way that whenever you upload a product image to whatever app or portal that you are using, you can have a microservice that listens to that change. That microservice will be built and running on Catalyst and it will read that image, generate the alt text and automatically update it in the marketplace's relevant field through the API. So you can completely automate this process instead of having a UI for this, but I have built a UI just for demonstration purposes. So I have uploaded an image and I'm clicking on generate alt text. And you can see that it is giving a uh an SEO optimized text that you can then copy and paste it wherever that you want. So the back end for this is also pretty simple, just a serverless function that is making an API call. But the only difference here of course is that I'm not using the standard LLM. I'm using a vision language model here. So that vision model, as I showed you before, can be accessed by clicking on Quick ML LLM serving and then selecting the vision model here. And the sample, I mean the way you need to send the data request data for this model is slightly different. So if you're using images, you need to send it as a basically a base64 encoded image and you will get a sample response, you know, back from the model as well. So these are three different quick use cases. The idea here is that I want to introduce you all to the different services that Catalyst has to use AI and serverless and back end and all of these together and hopefully this gives you some directions on where you can go next when it comes to implementing your own use cases and scenarios. Now there are a few more you know cases where you can use Catalyst AI capabilities and that is not specifically restricted to LLMs. So Catalyst also has a component called as Quick ML which allows you to build your own custom business intelligence models like your own recommendation models or your own say transaction fraud detection models provided you have your own data for it. So if you have the data, you can bring that data over to Quick ML and then you can basically use our builder to select which algorithms and data pre-processing that you want to do and you will basically get a built model and its endpoint available to you to use in your apps. So Quick ML uh is something that would deserve its own session specifically. So that is not the scope of this particular video. The other AI-based service that we have is Zia. I'm sure you all have heard of Zia. So we have a bunch of pre-built models that you can use here either in facial analytics, OCR, ID scanner, image moderation or text analysis. Say if you want to do sentiment analysis or keyword extraction. You can do that as well. So like I told you before um you know, contextual intelligence is

### [37:16](https://www.youtube.com/watch?v=7C88zn1JRi4&t=2236s) Zia – Context Intelligence Engine

very critical when you're trying to use LLM because if you don't use it, you may not get specialized output from the LLM. It's just going to be generic and not specific to your needs and use cases. And you know, our contextual engine called as Zia which I just showed you is specialized in executing singular build-defined tasks with precision. So in Catalyst we have integrated and provided support for all of these and you can basically use the Zia models as the foundation and use it for specific machine learning use cases and then do a bunch of LLM actions on top. Now the first use case that I would like to maybe talk about is vision. So when you are selling products it's essential that you maintain accurate images for those products from either the product's latest or the color and customers prefer not to buy products with incorrect, unrelatable or absent images. So with bad product images, there's a chance of you know, losing customers or having to invest a lot of time into correcting these images. So this image validation feature is a part of Zia's vision AI capabilities and that is exposed inside Catalyst as well. So it can do both classification and detection. So for classification, it considers the entire image for validation and matches it to the pattern learned through the training data. Say for example, you can use our AI capabilities in Catalyst to prevent someone from accidentally uploading a picture of say washing machine instead of a refrigerator. So you can define desired or undesired images and you know, based on Zia's validation for acceptable and unacceptable images respectively. Now for image detection, instead of assessing the entire image, Zia matches a part of an image to the pattern learned from the training. For example, a car must have a number plate. So if the number plate isn't detected, it's considered invalid and sent manual approval or rejection or whatever process that you could define through serverless functions or any other Catalyst components. — [snorts] — So, let's say you are a car reseller and you need to ensure that the uh cars that you sell do not contain dents or paint scripts. So, you upload the images of the cars that you are reselling and uh create an approval process where only images with no dents and uh paint scripts are approved. So, let's say one of the cars has a dent that reflects in the uploaded image and um you know, the validation will fail and the image will be sent for uh manual approval or whatever it is that your specific process is.

### [40:32](https://www.youtube.com/watch?v=7C88zn1JRi4&t=2432s) GenAI Actions

So, on top of the vision model, you can do a bunch of generative AI actions using the Catalyst LLM feature. So, one is, of course, automated reporting. You can uh use LLM to basically create a detailed report of what the model vision model found and send it to the moderation team. Or you can also draft a response back to the person who uploaded the image in your uh you know, reselling platform saying why it is against the platform's guidelines and why it cannot be accepted. And it is also possible to push this knowledge into the uh KB uh because uh you know, the LLM can create the appropriate KB entry uh by itself. So, this can either be used for uh you know, internal support purposes or depending on the information that's there in the KB, it can be uh exposed outside as well.

### [41:28](https://www.youtube.com/watch?v=7C88zn1JRi4&t=2488s) Sentiment Analysis

Uh next up is uh sentiment analysis. So, let's say um a company may receive a large number of emails every day, but the ones that have a negative tone usually require more immediate attention over the others. But to identify the emails that need to be prioritized or require immediate attention, you need to analyze your content carefully. So, the email subject line may not always indicate what the purpose of the email is and the amount of analysis when done manually is a very time-consuming affair. So, this is where uh the email uh text analysis and sentiment analysis models that we provide would come in. So, it groups your mails into various different categories, positive, negative, neutral, and uh emails with a happy tone are grouped under positive. So, I'm just using emails as an example, it can be any text. And those with an unhappy uh or neutral tone are grouped accordingly. So, let's say um you have uh received an email from a customer saying that she had reached out to support regarding our product, but it's been a long time since she got a response from you. So, the sentiment of this uh email is negative since the customer is unhappy with your support. Hence, you should prioritize this customer and quickly attend to her queries. So, uh this uh sentiment analysis can now be extended with a bunch of generative AI actions on top of it. Say, you can uh create an actionable summary of tasks that a support person can do uh and also a summary of the entire uh conversation thread with that particular customer. And uh it can also suggest possible solutions for this customer uh based on uh information that is there in the knowledge base. And then uh if you want to take this a step further, you can also do automated uh response drafting. So, uh you can have the LLM create a draft response that the support person can manually verify and then uh send it over to the customer to uh you know, save time. — [snorts]

### [43:52](https://www.youtube.com/watch?v=7C88zn1JRi4&t=2632s) Smart Recommendations

— Next is the uh recommendation capabilities. So, uh our models are able to identify and analyze customer data like uh purchase details, interests, requirements, and behavioral patterns in order to suggest the most relevant products and services. So, in formulating uh recommendations, the AI assistant uh or the model also compares customers' behavioral patterns to the other customers with similar attributes. So, uh you can create your own recommendation models according to your various business needs using QuickML and then you can set up notifications or uh anything else to send recommendation to your reps. So, they can use these to uh provide your customers with the uh you know, the right products and services and also improve cross-selling. So, furthermore, you can also send notifications uh every day or every week using things like the cron job or the mail components that Catalyst has and uh you can refine them based on custom criteria. Now, let's say uh you are running uh you know, an uh e-commerce business billing selling various TVs. One of your customers is looking to buy a TV and has purchased a significant number of other electronic items from a particular brand. So, you can have this custom model scan through the existing customer data and have it recommend television models from that particular brand. So, now that the model has given you the uh recommendation data, you can now use uh LLMs to do things like uh send automated uh recommendation emails to the customer say every month or every week or whatever time frame that you want. Uh or you can also integrate this into uh our uh LLM chatbot. So, we provide that uh feature as well in Catalyst. So, that someone who's coming and uh chatting with this bot uh can ask for what products that they can buy and the bot can then recommend uh the TV from that particular brand to them. So, yeah, these are uh some of the uh use cases that you can uh build combining Catalyst uh serverless and AI capabilities. And uh with that, we'll come to the end of this session.

---
*Источник: https://ekstraktznaniy.ru/video/46048*