CA Innovators Showcase AI Solutions for Audit and Reconciliation

life. So how CA can make it more effective and efficient in order to save time, right? Uh so we say that a 4-hour reconcilation can be done in 4 minutes. So it's like 10 times saved for a charted accountant. How you can probably use that for your more efficient and effective work and do more work and probably do more business, right? So while doing this, can you just talk about it and just discuss about it and you'll achieve that? No, you need to upskill yourself. So when I'm talking about upskill, it's not only about watching some content on YouTube or LinkedIn which you will just look we will kind of give a 30,000 overview of what AI can do. But what you really need is a ground reality on the ground how you will implement AI, how you will how that will make change in your life as a chartered accountant, how you'll use it in audits, how you doing in stacks. So we thought over this and we kind of uh come out with a learning uh program for you for charted accountants for us called Kyra. Now what is Kyra? So Kaira is a platform where in you get a access to the various sessions basically master classes we call it as and you can learn over there we divided into kind of three focus areas I would talk about that the focus area one how you use AI like copilot now everybody know chat jity claude that's that's not only the AI it's much beyond that so if you integrate copilot with excel word which most of you use how you can make your life easy. Not only that, it's not only about copilot or LLMs basically, it's beyond that like having your own AI agents. So, how many of you actually would have created your own AI agents? I think very few, right? Very few of you might have created. So, if you want to create your own AI agent, can you do that? Yes. So, that's what we actually try to try to show in our uh sessions. The second focus area which we have is basically you know now how to build your AI and versus there are tools which are available. So you can take more calculated decision whether you want to buy a tool or you want to spend your time and then kind of develop your own AI agent. And the third focus area or the sessions which we have over here is more related to AI. Round of applause. I'd like to invite first and foremost CAF Kotari. Please come and be comfortable with your laptop. Meanwhile, let me just share the journey of innovation challenge and the rules for the competition. So when we did AI hackathon yesterday, that was season 5 grand finale. But you'd be happy to note that this is the maiden edition of ICI AI innovation challenge. So this itself is an innovation of conducting the first edition. The proposal was floated by ICI inviting innovation using AI and there were about 30 proposals that came from chartered accountants across the country and then there was an esteemed jury who went through those proposals and shortlisted 12 out of those 30. Once we came at this stage where we had 12 chartered accountants proposals being shortlisted then there was an online presentation slot was given to each of those 12 and jury again evaluated this time the 12 people were presenting out of those 12 finally four are in the finals today and will be making this presentation as part of the AI innovation summit 2026 so that was the journey in terms of the rules first time. Each finalist has got 12 minutes to present their innovation. This will be followed by jury interaction. There will be a buzzer and a countdown timer that will guide you in terms of your resources with time. And we request every presenter to please respect time. And in terms of evaluation methodology, it is pretty democratized. The jury will evaluate but their score sheet carries 70% weightage. The remaining 30% is in the hands of each person who's sitting here. Now in the WhatsApp community group you must have received a Google form. So if you open the Google form you'll find name of the four finalist and alongside a scoring from one through five. One is the minimum, five is the maximum as you see these four innovators present their side of the story. Please do rank them and your response will be considered valid for submission only once you have ranked all four. So please do that because your vote matters. So that was about the rules and if there is any technical glitch and we lose time or any other such issue the jury's decision will be final. Okay. So let me introduce now CA Bareraf Kotari to you. Our first finalist who will present. He's a rank holder at the foundation level and has been drawn to technology and automation from his early years in the profession. And after two years with the finance company, he rose to group CFO with another company. So it was quite a transition and along the way he has closed several crossber merger and acquisition deals as well. He saw the opportunity with AI and that led him over the past 18 months to immerse himself in the field of artificial intelligence and full stack development not just as a spectator but as a builder and one more dimension that he brings with himself is that CAF Kotari is also a filmmaker. Go back five years in 2021 he directed a Gujarati film that was a biopic. So that's another very interesting creative dimension. All the best. Over to you. Thank you.

Thank you very much. Thank you.

I've been in practice for several years, 18 years with AI. I wanted something which could allow me to scale my practice, grow big. Crazy ambitions. I want to be big as as bigger as a big four. But I wanted an ability that I could scale the organization, empower every individual in my team with AI with guardrails. So that's where I started this journey of preparing this application. I'm going to skip this slide and get to how this thought process started. And as I worked on building this for me, building an application is like a building a car. You have to get a very good engine, very bad engine, it fails. So a very good engine which does hallucination control. And I'm going to give you an example during my live demonstration. Second is it should be very easy. You know historically applications have been everyone talks about you know implementation time 6 months implementation 12 months implementation time. I wanted an engine which automatically learns understands a user and so that the implementation time could be as low as half an hour. AI orchestration today uh AI there are a lot of opportunities. So I've spent a lot of time in building an architecture and engine and I'll show you a blueprint quick blueprint about that where depending on the prompt it decides which model to use what temperature how it should go step one step two loop. So the complete orchestration with the end end goal saying that it should give phenomenal results. An engine without features is not so great. So uh I wanted features like agentic AI. Can it do work for me and not just tell me what should I do? Can it help me establish SOPs in my company so that everyone follows the same set of checklist templates and AI powered uh can you help me with analysis which is where I broadly know what could be done. A client asks me, "Sir, can you tell me how will this impact me? And uh how do I ensure I have my clients and my team on the same platform? Safety is a must. It's not optional. So PII reduction, anonymization, uh we use enterprise APIs. Therefore, this the data that goes out is also covered by kind of agreements where the language models don't use that for the training. very powerful redaction model where you know it looks at all information and then automatically anonymizes it and then splits data into different LLMs once it comes back it hydrates the information so that for a user it's a seamless experience but in the back end there's a lot of safety and uh you know you would have if you if you Google around there was this popular thing which went out saying that LLMs could not figure out how many hours were there in strawberry and that's precisely the problem because LLM is a language model. You can't give a math problem. I mean if you give it it may work but sometimes it may make a mistake. So a deterministic path saying if there is math it should use a math function and not an LLM. And finally it should be something that we all want to flaunt about that look this is what I'm using. So great exteriors and exteriors is not just about you know a front-end screen. I looked at things like uh if I'm spending hours on it, I want my entire application in dark mode so that it doesn't stress my eyes a lot. I want something which is WCAJ compliant so that readability even people with certain uh challenges they are comfortable reading and fast intuitive UI UX all of those features three core uh principles. How can it help me increase revenue? Cost reduction is fine. Productivity is is important and it's a given. It's a must have. But can it help me increase revenue confidentiality is a must and hallucination control. This is the broad framework of so built-in 10 layers of safety. So the moment you have any request it looks at injection uh jailbreaking blocking consent malware. So it goes through the cycle looks at all safety measures. It comes to a decision point. If it needs a deterministic engine, use the math function. If it needs, you know, a language model to work, it uses multiple language models. So again, the language models are not hardcoded. These are through a site admin module. So today there could be three, four, five, tomorrow there could be 20. It then evaluates which one is the right one and then once the data is out after then it again rehydrates so that the user gets good quality report. It's built on a modern type- safe uh multi-tenant cloud platform. Front end typescript react u backend I've used posgress sql uh I use cloudflare for storing documents of kind of caching versel for kind of hosting clerk for authentication. So world-class architecture. The idea was can it tackle clients across the globe? Can it can this be