Name: Building the Foundation for AI: How Celigo Powers Reliable, AI-Ready Data
Uploaded: 2025-11-04T17:16:04.754Z
Duration: 56 min 43 s
Description: Building the Foundation for AI: How Celigo Powers Reliable, AI-Ready Data

Transcript for "Building the Foundation for AI: How Celigo Powers Reliable, AI-Ready Data": Hello, everyone. Yeah. See that Alexia is from Knoxville, Tennessee. So anyone from where are you, located today? I'm located in Frederick, Maryland Northeast, so not sure. I think it's a little bit early for, people from Pacific. Well, yeah, I think we got people from all around the world. See, Sao Paulo, Brazil, Philippines. Good to see people from, different peep regions of the the world. Okay. I guess let's let's start, because we have a lot of content to share with you today. And, yeah. So, thank you everyone for joining us. My name is Youssef Zouhairi. I'm one of the solution consultants here at Celigo, and, I will be joined with, Sandeep. Sandeep is the director of AI, operations and development. So this this this session is more for the builders and the technical people. So, and and Sandeep, so we we brought to you like, Sandeep is leading the team internally in Sunigo because he's the one, and his team is is building all the internal AI use cases that run-in the back, back end for, for Celigo Sandeep. as we said, this is, like, more about the practical part, how to build these AI use cases and what what can we do, to have, these AI, like, successful AI, projects, not only projects that are in theory, good for the company, but also, like, impractical, and and that can go to the to the production, site. Okay. So next, we will start with, what teams need from AI, and why many pilots pilot projects fail. And then we will show what's what is possible now, and how to build it, on integrator.io. After that, we will have a short demo, just to show you a couple of use cases that we've built. And we will try to get into the technical the to deep dive into the technical side of it. And please, with the during that, that part I mean, if you have any technical questions, just put those questions in the q and a section on the top right so that we can answer those questions later on when we get to to them. So one one simple theme will be for this session is that AI is is everywhere and, but not every AI project is successful. We think that Senego as a platform, integration platform as a service, is is is really, necessary for for a lot of companies to have successful AI projects. Because this gives you, like, things like the connectivity, readiness, and and and action. Okay. So let's start with, what teams, need from from AI. So most teams want, faster answers for their chatbots, fewer escalations when it comes to support, better customer and employee experience, and the ability to act on on on data, not just read it. And the gap isn't isn't really the vision because we have the vision. We all have ideas. It's more the of operations part, and this is why, like, someone like Sandeep is very important, for for a company like Celigo because they they take those ideas from different organizations and different, departments that we have, and they put them into the operations part and build, something that can go to the production. So, we will see how we can, like, the these teams, they not they they they need something that can connect to the right data, keep keep it clean and, and current, and then run-in, safe workflows end to end. So, let let's see what does good look like, really. So we picked, these three, because they show up in every successful AI project that we've seen with our customers no matter what the team is. So, everyone wants, like, faster answers for for a better experience, and that's the speed of the customers and internal users actually, like, feel. They they need to get the, things done quickly. The second point is basically time back for human experts. So you have a lot of, people in your own organizations with tons of skills, and you don't want them to be stuck in, the the repeat repeatable, like, the routine, dozy work type of tasks. And you want to give them time back to focus on, like, the edge cases that can make difference for your company. And the third point is the revenue execution gains. So AI that moves we we want AI that moves like for, like, for example, the sales team, the pipeline that, can give, like, more, provide retention, for customers when it comes to, to customer success and expand for the customers, for more use cases that that they can see, the the themselves, like, using your your tool or your, your product for. So these are, like, the outcomes leaders measure day day to day and the ones AI can support. And if we built built it right, this is going to be very, very helpful for any organization. So this is having said that, many of the pilots don't get there, and this is why. So, in reality, many organizations have fragmented data, disconnected workflows, weak governance, and manual optimization needed, to, to to to do a lot of things. And this is why so, in a recent, study, 88% of the projects, the pilot projects, are not going to the production. So and, they they never reach the production side. And 80 95%, lack really the ROI. So, it's not like a lack of, ideas as I mentioned before. It's more like a missing foundation. So we don't have the foundation needed, to make this work really for, for everyone. So what are the so what is this foundation? We are talking we keep talking about foundation. What what do we need as a foundation here? So, three steps to unlock the production, level. So we have first thing is the connectivity. We need to have AI that so we need to be able the AI need need to be able to reach, the right data. The second one is we need to be AI ready. It means that that data needs to be cleaned, needs to be current, needs to be consistent so that the AI can be used can use it, correctly and give, like, more, like, insightful, outcome from from the use of that data. The third one is the con contextual, aspect. So we don't want to to use, like, AI, like, currently, everyone uses, like, the GPT, for example. If you ask a question, you get the GPT answer, and the the models are really, really good right now, but you need to have the context of your business. You need to have something grounded in your business so that the results can be, relevant to what you are trying to build. And as Forrester, you see here Forrester, one of the, like, biggest analyst, firms mentions here is, like, even even when agents technology is ready, adoption fails, without these three aspects, the connectivity, the AI readiness, and the contextual aspect. And this is why integration matters, for the AI. So, in fact so AI plus, integration is where value shows up. So integration provides connectivity, orchestration, and we will see where we have every aspect of this in, like, something like in NYPES, like, Celigo. So connectivity, orchestration, security, and governance. AI transforms how we, we build and run and processes with these, four aspects. So another analyst, view, says that by by 2028, 70 almost 70% of the organization building multi LLM apps and AI agents will use integration platforms for data orchestration. And I think anyone who is really close to the AI, space right now will notice that we are moving from the LNMs, which are like ChartiPT and Propic, or like the systems, the different CRMs, the different ERPs to, like, use cases built on on data moving between all these, this type of systems from the so we we need to orchestrate data moving between the ERP, the CRM, the marketing tools, and then, using the the power of the the the power and the the big, knowledge that has, the these elements to basically bring the ROI, the benefits needed, for for the company. Okay. So with with that space, here is what this is what what is possible with, with, today to build today on on Celigo. So let's see what's, awesome, proven patterns that Celigo, ourselves and customers implemented today, like the cost the support automation, things like customer three sixty, back office document workflows, and, these are like live today in our, like, in in Citibank instances and, our customers' instances. So let's let's first, take a quick, platform review, or view like we have, on on one platform, we can connect multiple, applications. You see the connectivity side. So we have connectors to different different lot of tons of, applications, SaaS applications, on prem applications, databases that you can use. We also keep, the data AI ready, by using, like, the different LMS, the most powerful ones. And we have a lot of, like, connectors to, whether it's, you can see in the, in in the slides here, we have, like, OpenAI, on topic for cloud, Perplexity for, search, which is, like, very, very powerful today. And then we have Gemini. That's that is, like, gives you already, like, tons of tools to do. Like, I I heard that Gemini is very, very good when it comes to parsing, like, PDFs, for example. So every one of these models can be really good in one, aspect. Like, for example, cloud is more like for code generation. ChargeGPT is for, general, use of of AI. Gemini, as I as I said, it's more like for the parsing, images and the PDFs. And Perplexity is, of course, is the one of the leaders when it comes to search right now. And, so the all that can can can be contextualized by using your CRM data, as I said, your support tickets, system or your ERP to to bring the the value needed, here. So, let's let's see. We have, like, I picked for you, like, four concrete examples that we are we haven't already, like, live today just to give you, like, some inspiration. And then after that, we will deep dive on into what is needed technically to build these use cases. So, the first one, is the customer 360. So in Slack, an account manager can ask things like, what is, like give me, like, details about Acme as as a company. So the workflow pulls data from Salesforce. You can see Salesforce. You can get, like, the product usage from, Snowflake, as a data warehouse. You can get the support tickets opened by that customer, that accounts, and you can get, like, the calls from Gong or any other, like Zoom, for example, transcripts. And you can provide everything to the AI with the orchestration done by Celigo. And at the end, you generate, like, a PDF that can be, like, a a good preparation for that accounts manager, to have a discussion with that customer or also, like, have, like, a clear risk or expansion view, for that account. And and that's that PDF is is usually, like, shareable that you can share internally or you can give to your executive leader to, to have, like, a very, very good discussion with that customer. So the second use case is, support leads, asks for, this for this customer support AI agent. Support lead can ask for the best resolution email draft for a real ticket, and this agent is, reach can retrieve the policies from your company. Like, you all the documents that you have in your company, you get, like, the the policies, the knowledge base that you that you have in your support organization, every past fixes or, like, bugs that were resolved, and drops for you, like, a grounded reply for your customer. It's gonna be like a a a start. Like, for that support rep, they can read the the the drafted, reply. And then from there, it can go either, like, adjust it or or basically just share it with the with the customer. So the outcome here is, like, something like you can lower the MTTR, you can have consistent, answer, you'll have, like, a deflection of repeats questions. So every time that you have, like, the same questions from the same customers instead of, like, every time I'm not saying, like, wasting time, but spending more time on finding the the answers for the same questions. You can automate that by with the help of the AI to get those questions answered quickly. So the next one is, is a very, useful one used by our internal team, the AR team, accounts receivable team. So, again, we noticed some patterns like, there are questions. Our AR team, they receive emails for the same, type of, requests. So they whether, like, someone like a customer is asking for the w nine or, like, invoices, or can you give me, like, a statement for my for my company, between the, let's say, the November 1 to the November? Those those inquiries are, like, simple, but, the AR team needs to spend a lot of time and efforts to reply to these emails. So what's happening here with this agent is that, you get the email, you read the emails, right, from your generic inbox email for the AR team, and you read the content of the email. And with the help again of Celigo, you can orchestrate on the AI. You can you can read the content and know what is the action, the next action to do. And based on that action, let's say that this email is is, for example, requesting, like, the invoice or statement for the previous month. So you can just, with the with the on an orchestration, apply and do multiple, actions. Like, for example, retrieving that, that statement from your ERP and then, and then build in, the PDF for that statement and then send and then write you write, like, a draft email, for that AR team member, and you they will have it as a draft email that they can review. This is where we it's very important to have that human in the loop and have someone, in the loop to review the what the AI is producing. So they can just review the email, see that everything is okay, and then send that email to, to whoever is sending the request or, trying to get the answer. So the, the next one is, is also related to the financial team. And this is, this is the expense report approval agent. So, again, same thing. The financial team gets a lot of so when when there's, like, a company with a big sales team, for example, usually, they go to events, they go to, they they have, like, a lot of expense reports, to be approved. So with this agent, you auto approve you you auto approve the expenses when those expenses match the policies that are, drafted by the company. So, once they are compliant, they will be auto approved. If something is wrong, for edge cases, they are flagged and a person will receive the those those expenses via email that they can review and and see what's happening in in that use case. So we in in Sadigo, again, we are using this internally, and it's, it's a huge gain of time, for our financial team. So let's talk about how we can build, this use case, today. So this is, being the main point here. So, the first principle so we will see what are the technologies, the patterns, and the the architectures that can be for, like, someone who is building flows and building, automations in Celigo and Integrator IO. What how they can what are the main concepts needed, for this type of use cases? So, the first concept and the first principle is, in my opinion, the context engineering. So, prompts matter. Of course, you need to be able to, write a good prompt to the LLM, to the AI so that you can get a good outcome, but it's not, but it's not all what matters here. So, context engineering is what makes AI, the dependable. It sends the right info, in the right structure at the right moments in the flow. So you grab data from different systems. You need to have the right format of that data, and you will have you need to have it in the right time, with the right amount of information needed. You don't need to overwhelm the LNM with too much information, and you you you can't, like, just provide the LNM, like, like, fewer information and then expect that the l n m will will give you, like, the the right answer. That means, simply that you need to connect trusted systems, clean and shape, the data and chunk log documents. For example, you have documents in your HR team, and you have those documents that cannot be ingested as is, but needs to be chunked into small chunks that can be, where we can retrieve the information needed and and share it with the LMM. You need also to have, like, to feed the models with the model ready JSON, CSV, XML, all these these these structures. So a common way to deliver this context is something that we call RAC. And RAC is, acronym for retrieval augmented generation. And retrieval augmented generation is is a pattern used, in AI, and it's it just helps to fetch the right facts, from your own systems first, then have the model generate an answer grounded, into that context, the context that we just, discussed few minutes ago, with citations. So we, we provide the reference. Everything that we provide as an answer using the AI, we need to provide reference and cite where we got this information from. And drag is a very good pattern to explain and to provide this information to whoever is user. It gives, like, more credibility to the user. It gives, like, like, trust when it comes to what is the source of the the this information. So the key concepts, of the direct are, first of all, we need to ingest. So ingest trusted sources via connectors. So you ingest the information. Then we clean and chunk that information into small chunks that will then be embedded. So an embedding is basically you take a text text and with the help of AI models, we transform you transform that text into a vector. Like, you tokenize it into a vector. Something like a model like, OpenAI, can can create, like, a vector of 1,536 or thirty thirty two or 36 dimensions. So you take no matter what how big is the text, you can represent it into a vector of numerical values, and that vector will be stored then in a vector database with so you get the text, you get the vector, and you get the metadata. When did you get ingest ingested this? What where did you ingest the test from? What is the source? So you you store everything in a tie it's very special type of database. It's called vector database. And that vector database has a very interesting feature that makes everything here possible. It's the fact that you can do something called similarity search. So instead of trying to find values by trying to get the actual keywords, like the words that you are doing the search with, the similarity search can give you the possibility to to, to search by the meaning. And this is very helpful. So so you see here the, once you you generate the, once you you invest the the information and you store it, now you in you assemble everything into, like, a JSON, and you provide that to the LMM, to the OpenAI, for example, the the AI that you are using. And then the LMM will take that and then will generate an answer and will generate the citation where did they get the information from. So this is this is the main concept behind a lot of use cases that you see today in the market when it comes to AI. RAC is is a is a, like, central piece to this because it gives you this possibility to retrieve information that's the the LMM or the AI needs by just doing similarity search. Okay. So the third part and, like, I mean, the next one needed the next concept needed is the, guardrails, and it's very important. This is this is one of the main, concepts that, makes an AI use cases an AI use case or like a project go to production or fail and stay as a in a development mode. So with GuardTrails, if you are like, enterprise level, you you will need to, put into, into your AI, application thinks to to do the input moderation. You don't want to accept everything. You need to to to scope check. So, for example, if someone is asking about, so you have, like, let's say, you have, like, an agent related to customers, to customer support. So that agent, is supposed to receive only things about support. Right? So a ticket, something like technical that needs to we need, like, an answer to. If someone is asking about, I'd like, like, accounts, for example, can you give me, give me, like, the accounts that you have with the name x y z? This is an information that we should not share with the the requester. This is where the input moderation or, like, the input validation is very important. The second one is the output validation. So now after you get that response from the, you do the orchestration, you get the response from the LLM or or from the the AI, the you you need to to validate that output. One of the technique used here is the, what we call an LLM judge. It's at the end of your flow, you add, like, a step with the AI that is independent from everything that's, was implemented in that flow. And that's that step will have only the, one task, which is, like, to judge if the answer, is, is compliant with a set of policies that you have. Like, for example, instructions, you don't share PII, you don't share information about, about accounts, you don't so you you put some a set of instructions and policies, and that output validation is very important there, to to make sure that your your AI application is is safe to to be used. So the the the third point is that data governance. And this this one is inherited in our case with Celigo, from the platform. So by using Seligu as an iPaaS that will, will, basically, be the engine for this AI orchestration, you inherit by default the data governance. You get the airbag, permissions, the SSO, the encryption, address, and it charges it. You have the audit logs for anything happening with your AI calls, like the API calls and everything happening in the platform, you have multiple environments to, to do the the the developments, the QA, and the production. So this is inherited by default with the infrastructure of the platform. So a fourth and last point here for the guardrails is the testing and rollout. So it's very important to test, and and the testing can be done, with the test mode in in, like, in in the flow builder in Celigo. You can have, like, feedback loop when it comes to building, like, a human in the loop by by, integrating and having, like, the your your AI use case connected to some something like Slack or Microsoft Teams to always have, like, some type of login sent to, the users. And and you will have, like, the metrics needed from the platform to to make sure that you have everything safe when when it comes to using this, this AI, agency. Okay. So let's see this in action now. I will share my screen. Sandeep? So let's go and look at the first use case, which is the customer, 360. So how does the customer 360 work? It's very simple, fairly simple. We are orchestration we are orchestrating, data between, like, different systems. So for this demo, I will do one thing. So to be able to, our front end or how to interact with the agents, that you see here in the back end, we are using Slack. And we have something called, like, Slackbot, and we are implemented we are implementing the Slackbot, slash command. And for the slash command, we are using the slash, account summary. So, with account summary, I can just give it the name of the company. So Acme, that's the example that we saw. So once you we we, provide the name of the, the the the company or the account, first thing that the the bot will do is will try to find what are the accounts in our CRM in Sirigo with the name Acme. And found like the first one, which is Acme Triplettier, see it's like a customer closed one. It's our customer, some information about that customer, and then I will pick that customer for my demo. So choose now what will happen in the back end, that will trigger, this flow. thing that we are doing is we are getting the list of Gong conversations, for that customer. We are going to get all the transcripts related to that customer, and then we are going to get, using the organization, the tickets and the comments, everything related to support where we interacted with that customer. Then we will get the this from Snowflake, how that customer is using our platform. This is something that we store in our data data warehouse. And then we have two AIs that we are using, OpenAI and Gemini to do the analysis of the transcripts and everything that we provided. We generate then a PDF. Right? And then we once we generate the PDF, we create the doc in Google Drive, upload the PDF, and we send back a response, to to to the link. So yeah. You see that when we we we, selected that, account, now we got the documents from Google Docs. So if I click on this one, I will receive a very nice, PDF, that's like the account manager or like you the executive leader in your company can can get this one. It's generated today. It has all the information about the customer, like a high overview of who is the customer, the profile, thinks about the customer business overview, what we did with that customer, the key stakeholders, with their, email addresses and everything. And then we get, like, to, financial, data with that customer, how many users they use the, the the the platform. And then we get to things where we whether, like, very where we can flag the risk. Red flags, for that customer and and how many support tickets that are still open right now and a quick summary of every one of these support tickets. And then for the expansion part, so if one of our account manager is, like, trying to have a discussion with that customer to see where we can, like, bring value to that customer, more than what they get with the with our platform, they can have strategic discussion points. And this is it gives give them, like, quickly, everything that that is needed for this discussion. So you can see that with one, AI agent, the, customer three sixty, we generate something like a PDF like this one where we can bring quickly, all the points related to that customer, whether it's it's for the, expansion parts or the risk or even, like, just to have, a high level overview of that customer for like any executive maybe, a discussion, that we can have with, with with that user. K. So if you have any questions on how this is implemented, please, feel free to, put the questions in the q and a. So you see that, we are using a lot of, tools here. So we are using transformation because we want to transform the data that coming from Slack and get only the the one that will be needed to run our flow. We are using branching to branch out the different, cases on, on how to handle the that slash command, from, like, overriding the message to finding the, the account and basically, providing, like, the the the choice that that choose which accounts do you want to, to select. And the main you the main, branch where we are really doing, like, the, creating that summary by orchestrating, different steps here and getting like at at the end from with the help of the AI, some some, like, documents that can be used, for for this. Okay. So the second point that I want to show you, the second use case is the AR one. This one is, if you remember the one where, the accounts, the accounts receivable, the accounting team, they need to, to auto respond to some usual simple tasks or simple asks, from different customers. So customers asking about, let's say, the statement. They will they will send an email to a generic email, and then we have, like, this, this step where we read the emails, we read the content of the emails, and then we, we we provide we extract from the emails every inform every, information needed to build like an answer. We extract information related to the customer and then we go like NetSuite, we fetch all the information related to that customer. And then based on the content of the email, we categorize that email. What do we want to do with that email? What is the action, the action requested by the whoever is, sending that email? So we have here a few actions. Send back the bank details, update contact details, send invoices, send account statements, send w nine forms. Those are like if you ask anyone from the accounting team, those are like things that are very simple. It's like, basically, you get the email, you try to do it. It's always it's always going to be the same way. It's just that it takes a lot of time. So you see if you see, like, something like this one statement I I mean, if you are doing it manually, you would have to grab all the information, read the the the the the customer and get the details of the customer from NetSuite and then, for example, on the ERP. And then go to your ERP and go wherever you are retrieving the statements and try to extract that statement and drop an email, for that user. It might take maybe like twenty minutes to, to do that, thirty minutes. With this flow, it's automatic. And the once the the statement is created, the draft is create like your draft email is created and then you you move that email to the process draft, label in in, in Gmail here. Same thing for this one. This is like a very interesting one, updates contact details. The company is sending you like, a customer is sending you, well, we changed our, address from, I don't know, from Maryland to California, for example. And, and and they give you, like, the address the their new address and maybe the name, the new name of the company. So, so someone from the accounting, again, needs to go, retrieve this information, go to the ERP, and change everything there. So here, with the help and the power of AI, you can read the email, understand what's happening, categorize what action, and then retrieve all the information needed, get for example, here, once we want to, update the contact details, you get, like, the primary, account details, from from everything that we retrieved from the account from the customer and from the email, and then we start updating. You see here we, we we set the primary billing, for example, and set the primary so we do some updates on that, contact emails, details. And then we draft an email, nice one, at the end, and we move it to the draft labels. It's an automation. It's using AI. It's definitely providing like ROI here, return on investments, because you have like, a lot of gain when it comes to time and effort. And the accounting team can focus on the main things where it needs like a lot of analysis, and, like a lot of human life type of tasks. So, I want to touch on one last, example. And this is not, it's not part of the, it's not part of the the the use cases that I showed you, slides, but it shows you, the guardrail spot. So if I look at, you know, those are like, by the way, our internal, use cases. And if we look at customer insights, so this one. This is this is, an AI workflow, that provides a lot of agentic features, where we retrieve, data from from structured we we retrieve, like, insights about customers from structured data that we have in databases or data warehouses. So in our case here, we are using Snowflake. So what I want to show you is, the the guardrails. So let's say someone is asking a question to this, to this agent, agentic workflow. They can ask something where they will say, well, can you delete accounts, a b c? Obviously, this is an action that we don't want to allow, and we don't want to have them, to have them, like, the have the possibility to for them to do it. So what we do is we build the guardrails, in the branching part by saying like every query generated that starts with insert, updates, deletes, drop, truncate, anything that will modify or alter our database, we just take it to this branch where we say, no modification is allowed. So here we are doing like the input validation. Same thing. Someone is asking about this is like an AI, customer insights. It's more like a workflow, agent workflow to answer the customers. Someone is asking about, I don't know. So, something that is not related with customer. So with that, so in the age the prompt of the agents, we are prompting in a way that if it's not related, just give us something. And then we using the branch and we take that in a branch where we say, this the AI does not know. So we refuse to answer any question that is not related with the topic where we are for what we are using this this, agentic flow. We we keep doing that many times like for example here, it's the error handling part. So or exception handling. So if if if an error happens, happens during the execution of that database, we take that error and we take it back to the to the requester, so that they can know exactly, what's what's happened during the execution of that flow. Okay. So, I think so. The the other thing, maybe it's the the the testing And and it's it's basically when you, disable a flow, you have the possibility to run it into test mode. You see that? The test mode. So test mode can give you a lot of information about what's happening there and during the the the build phase. Okay. So let's go back to, our presentation. Yes. Okay. So key takeaways. What did we see, and what what what are the tools needed, to build this type of, agentic workflows? We we we need things like, connectivity. So and Celigo provides connectors, prebuilt connectors, more than 1,000 prebuilt connectors. And even for the SaaS applications that, that you, you don't have like a pre built connector in the platform, you still have universal connectors to connect to HTTP like REST APIs for databases, gDBC, SFTP for the, file file, servers, GraphQL if needed, webhooks if you need to do like real time things like the ones that I just showed you right now with Slack. Connectivity is very important. Second thing, so the Flow Builder, with, with all the features in a Flow Builder from the transformation, the the the possibility to to, to use, scripts. And those scripts basically, for those advanced age, cases are very important because it gives you, like, more power than, like, just a note. So the platform is very Celigo's platform is very comprehensive when it comes to no code, low code. But sometimes, and we all know that, you have that advanced use case where it's just like it needs to be done in a code. So we have that in hooks so you can use them, and that will give you the possibility to add like metadata, build like, schemas that you need, define what type of structure do you want to build for that JSON. So the orchestration is done also at the platform level, with with the flow builder. So you do the orchestration. You can also do orchestration between, not only the SaaS applications that you have in a flow, but also by calling, like, APIs that you built in the platform. And this is another, like, level where it makes the synchronous, it's usually used when you need, like, a synchronous processing of the data. So, you you you build whatever you want as an API, and you do the orchestration in a flow by calling that API. You have the possibility to do to, leverage things like lookup cache. You can leverage things like, like like the the autonomous recovery and the error management layer that we have in the platform. Makes it, like, easier for you to troubleshoot and debug, those agentic workflows that you are building. LNMs and vector integrations are some, like, governed and and, like, we have tons of these, connectors for all the LLMs needed, in the platform. So you have, access to OpenAI, Entropic, Gemini. You have, like, for the vector database, things like, Pinecone or, PostgreSQL with the vector p g. And I I I know there's a lot of people don't know about this, but you don't need to use PineCoral on on, like, a saascom like, a SaaS product to to do the, the similarity search. You can use it in a PostgreSQL by adding just one, it's, it's a plug it's not a plug in, but, it's a module that you can add to the PostgreSQL. It's called vector p g, and you can do the similarity search in your PostgreSQL, instead of using, like, something like PrimeOne. And the third one is basically builder, UX, helper. We have, like we we we have, like we talked today about AI use cases and how you can build on the platform, but there are also tons of features, AI features embedded in the platform, for the co as a code assistance. We talked about the scripting. Code assistance can can generate for you, with the help of AI, scripts that you need. You can do, like, auto mapping. You can do the generation. So it makes makes the experience of building this AI, agentic workflows really easy for, for the users. so, I think those are, like, the main take, takeaways from from this, from this, from this call. Sure if we have some questions, maybe before we get to the questions, just want to, remind everyone that tomorrow is the last connective roadshow, and it will be in Atlanta. And, and you can join, actually, you can join this, the the AI summit that we will have tomorrow. I think it's 01:30, eastern time zone to 03:30 eastern time zone where, a lot of, like, insights about the AI will be shared. And you can you can join that, in in the, in in the, online, live online, by by just connecting to, I think we have it in the chat. Right? Yeah. So Alexia just put the the link to, to join the, the the AI summit tomorrow. Okay. Q and a. So let's see if we have any questions. Do we have any questions? I think I see some. Will we get a link for the recording? Yes. We do. What other other questions? How do we keep it safe, reduce hallucination and product PII, and production, ready? So, yeah. This is a very good question. So context in context engineer that we mentioned and the guardrails are very important for to answer this. You you will have the possibility to, you will have to the possibility to build the the guardrails that's LLL and judge at the end of your flow, that can detect any PII before you send it to whoever is asking the question. The hallucination can be reduced also with, with providing the context, the right context for the LLM, for the AI to answer the question. So, if you provide tons of information, so let's say someone is asking about, some of things specific for your, in your company and you provide all your documentation to the LNN, the hallucination would be high. That's for sure. Because you don't provide, like, the the right context. This is why, I think it's it's very important to keep in mind that context engineering, reduces the hallucination and to protect the from the, PII, we can use the guardrails. So the second one, another another question I think is, from Eric. Does Celigo have the same ability to interact with agents via Microsoft Teams rather than, than Slack? Yes. So this is the question. And, yes, we can connect. So we are platform agnostic, Eric. That's the good news. We don't care if you are using Slack or Microsoft Teams. We don't care if you are using OpenAI, Anthropic, or Gemini. We are platform agnostic. You can use Microsoft Teams if needed instead of Slack. And we have customers building currently with Microsoft Teams, and, we have already, like, something in the production for other customers who built the same type of chatbots, using Microsoft Teams. So Mary's asking, will this tech be available? Yes. I think we we will do. Alexian will make sure to, send this to everyone. What does a production grade rack and complex engineering build, look like, you know, on Integrator IO? I think it looks like what we just said. It will it will give you, like, a lot of, let me see. So, let me see. So for the direct yeah. So we, we I'm I'm looking at my, my slides just to see, if we answered this. So yeah. Yes. So production grade, RAG is is usually it's repeatable. It needs to ingest your data, transform your data, index it, retrieve it, generate the answer, and then observe, by logging the feedback of the of your customers. And you can implement that using the flow builder and the HTTP connectors, the connectors, your LLM of choice as I as I mentioned before. Let's see if we have any other question last question. If Zendesk, Salesforce, and others already have AI, why put Zendesk in the loop? This is a great question. Again, so every platform out there right now has their own AI. Every platform. The problem is that the data is is fragmented between and siloed in different places. Your CRM can act with their AI, product on the data that the CRM has. The same thing for the support. The same thing for the ERP. Celigo is a layer that can orchestrate between all these systems. You can take data from the CRM, add some data from the support. Like the customers three sixty, it's a very good example because you are spanning over, like, different systems and different AIs. And I'm sure if you are asking, like, for example, the CRM, about, like, just that account, we'll give you, like, very nice information. But you will not know what's happening in the in the support part. If you ask the support system about the tickets, it will give you, like, a detailed report. But getting all this information in one place, that that's the the challenge. And Celigo will help you with that because we that's what we do. It's an integration platform as a service, and all the analysts out there, they are saying the same thing. Those AI agents will live in, data integration platforms like like Celigo. Okay. I think I got over, like, all the questions. Please, if you have any other questions, let let us know. You can contact us, and we can, schedule some demo. If you are interested in to see, like a specific use case from the ones that we, we discussed. And, yeah. Hopefully, that was helpful for you, and thank you for joining joining us in this session. Thank you very much.