AI in ICAI Video Playlist

Testment for today. CA Panav Vas G Panav G over to you. Your time starts now. >> Just am I am I audible right? My screen is visible. I have some network issue right now. Here >> no is visible but uh yes sorry it is okay. Please continue. Okay. Okay. So, uh good evening everyone. I would like to thank Ici to giving me this opportunity. A few weeks ago, I attended the FFTD batch and one thought could not let me sleep at the night. I grew up up around this profession being my grandfather being practicing over more than 50 years and I kept keep asking myself the same question that why is finding truth in the client books is still so hard and why only few of us can ch do that so here I set out to change that and that's where foreign shastra AI comes in yeah so you already know how much how many types of data we handle we handle messy mult formatted data spread across various patterns uh and generic generic AI right now is very unsafe not very but we don't know where our data is being processed or stored so our work today is slow leaky and very hard to defend so here what it my app comes it's a co-pilot not a complete solution but it's masked by design it will identify directly indirectly and contextual identifiers of identifiers of the person before uploading into the AI So that only codes and counts ever leave the system. In the proud I am very excited to share that it my app does uh does not simply just flag you the items for every finding in it uh writes in plain words what it does and why it matters. So a CA who have never done a work in foreign sick can fully understand it and grow their practice in the way they could never. Recently I have tried this app on an engagement also and it worked very well and they were very happy saying that let me show you what did I mean by the unstructured messy data. So you can clearly see that the Excel data here is very unstructured and also look at the line items there are more than 15,000 line items are there. Addition to that you can see that there is you have to mute. >> This is structured data but you can see the line items here. Uh there are more than 10,000 10,000 line items here. So coming to the solution here I built this exe app which is called foreign shastrai. I have built an .exe file and I gave it a nice loading screen while it loads so that it gives a nice impression. So this is how the loading screen looks like. Meanwhile it's loading. Let me tell you about the technology step. So I've used the flask library to host it on the local host. Then I have made the front end on UIUX using HTML and JavaScript CSS. The back end it's use is Python. The database is used it's XQite and I have integrated Gemini API key here. And also it has AES 256 encryption. Now coming to the app, this is how the main dashboard looks like. You can see that nothing is selected. But now as we can see here you can see no client is selected. But this is where we can create a client and here we can assign the engagement. So let me select a client here. Suppose I select Nang textile. So in no time you can see that narang textile is selected here and the mission control is still empty but I have I have sorry I have made all this data I have analyzed all this data right now which we can put in the akar here aar I'm I will show you in the jury round if you permit me now right now as time constraint I'm loading the data directly it will take some time so while loading it I I'm saying that it can analyze all these type of data and many more too but I have confirmed this one for Now it has loaded more than you can see 91,94 transaction it have analyzed. Now I just go to the labs here. It has one 11 tabs along with this four new features. When I run the battery, it will run the 104 forensic tools itself. And across the 15 sorry 16 different different features itself for the foreign tool and as it runs and it will change the mission control into completely amazing new dashboard here. It will take some time as it's processing heavy data. Uh so just uh spare spare me some time and it will generate us the data. Yeah, it's done. So it has run the battery and now look at the mission control. We have executive relay here transactions here and I I also want to show you the dark mode here. So I also have a dark mode which I can you can see the flag by severity risky parties data held by register money under the lenses and many more structures structural way like this. So now coming back to the labs. So suppose we want to run a benford rule. So this is anka parika where we can do the benford analysis. So is it gave me the benford of the journal here. If I want to do the for purchase and in no time it gave me a purchase. These all are safe. So it gave me the safe here. Suppose I want to run an evidence here. So here is the evidence card and working paper. These two are linked. Suppose I create any working paper. Right now I'm putting random data. Just I save it. I go to evidence card and look it gave me the evidence card. I can open it and I can see it from here directly. I can load it. It also gave me the exception part. Now I can filter it on what type of it's open all anything I can select directly from here. And if the transactions are critical I can directly load working papers and card from here. The same way this is the working paper feature where we can load the working paper by severity directly into the tool. Now the labs this is the chalanka. This is the banish manipulation loss. Suppose this is navang textile for 2526. So it gave me the analysis. It also gave me what I meant that even layman even ca can understand. So it gave me the analysis of what it has done. Suppose I change the year and compute. Look, it also changes the line like what analysis it's actually giving. What does it mean of the answer it gave? Same way parika vulan that is rsf pratir. This is very important. This where I found the most important and critical findings. Here you can see the names. I can download the excel directly. I have already downloaded it. So it shows the result something like that. So here you can see the party ids are different and the names are different. I'll be I'll be going bit fast because I have to cover a lot here. So if suppose I want to check the potential connection, it completely filter me out the potential connection. The parties are different but Indrani processor and indanal printers private limited. So there might be a potential connection. But if I want to see the exact duplicate then I just click here and in no time the parties ID are different but names are same. So it gave me the exact duplicate. So this is how uh we can match the name and mail right now but I also have in mind about the location and mobile number too. Now same way we have this Chayan that is sampling studio, Anupath ratio analysis, Kalanukram that is time sequence lab, anomaly tab, mullia and shabda. I have all these features as far as let me show you the shabda. So it gave me this narration is being repeated this much time. What are the suspicious keyword enter in the voucher? What are the short narration on high value? It gave me the complete analysis that way. Now we have nam yendra. This is computer assisted audit techniques rule which gave me the complete computer assisted techniques here. Now manus this is very interesting. This is the behavior and psychological foreign. So we have fraud triangle or fraud diamond in the foreign six suppose let me pick the prias. So it gave me the complete you can see the pressure opportunity rationalization capability. So it gave me the complete analysis of this and this way it also gave me the top parties here list. Suppose I want to run a dossier or any party. Suppose this is ro suppose I click on here it build me a complete dossier here and it gave me the identity and where it held. Now if I want to generate an AI note first it will mask the complete data used by AI and it will generate a result something like that. So while identifying the party we can know what to keep in mind. So that was the dossier. Now >> one of the last two minutes >> yes in bandan we have this uh network graph which also shows that 10,19 entities linked are 51 relationship this connection node with two links. So, so also it gave me the complete analysis of that this kala is chronology where it gave me the total where the flag transactions are raised. I have showed you evidence card working papers and report studio. Suppose I want to generate a gap assessment report. So in no time I get gap assessment report. I also downloaded because it will take some time. So this is how it generates a gap assessment report in the firm data red along with the placeholders too like this. So it it looks very nice. Now sutra this is very important this is a fivestep agentic AI which runs this planner executor reviewer drafter and sentinel so as I run this workflow it will generate me uh results something like this so it whatever it will mask the data and whatever my app was not able to my rulebased app was not able to answer me it will analyze that like it's fictitious procurement vendor kickback revenue overstatement roundtpping asset misappropriation etc now uh I have this dishi which is a public screener which uses screener to generate result and I have also kept a dummy result. So this is way it generates a result of any public listed company like that we can analyze it directly. I have this GAN course which is a knowledge portal which gives all the important information here like this and if you want to do any real-time analysis it runs the API key Gemini API key and answer generate answer with confidence score. If confidence score is more than 90% then only it will generate us the result here. So that was the gan push and I have to show this user manual. So any person who want to know about how to use this app and what is this about he can read it in English Gujarati as well as Hindi. So that was the foreign shastra and I'm open for the questions now. Thank you so much. Uh very nice uh tool very interesting >> very nice tool. I really like the names actually to be very honest. It's something which is fresh and new and we don't often see it. I think Sumits will also agree on the same points that the names are very fresh and >> can you just tell me the source of the information like you have got me a very detailed report and frankly speaking in less than 10 minutes it is impossible that we verify whether it is accurate or not but it seems to be very detailed to be very honest but can you tell me like what is the source document? How have you uh entered the data and how have you built the application? I have first given the foreign standards book and all the books and information I have about the foreign six. I have completely made an analysis of those books I find those on online and I made that complete summary also and I made the uh overall structure of this book too. Then I give I made a completely different project on claude to give it the proper identity. Now based on it I have generated I have structure what I do what how I make application like in the cloud I brainstorm and discuss idea with it. So what I want to do then I'll create the P file as well as I have that's 11 more file structure I'll show you. So uh these all files I generated you can see my screen right so all these file I generated all these are markdown files I generated these files using the clude and then I'll give the file access to the cloud code as well as the codeex then I designed the complete UIUX using the codeex and I built the back end using the cloud code. So and whenever the troubleshootings are there I just kept doing there are more features to this tool too but in 10 minutes I was I'm not able to show you but that's what the overall structure how I made it. No. Very nice. >> What database? >> Could you show me the source? >> Sure. These all files I uploaded as a source. And I have some other observation too which I have brainstormed here in the chat itself. No, I meant the source in your application. >> Okay. Okay. Sure, sure, sure, sure. Uh I have created this application built itself in cell here all uh sorry uh uh not this in the >> your application you have to upload a file, right? >> Oh, okay. Okay. Uh you may Okay. Yes. Uh yes, this one. Yeah, this one, right? >> Mhm. So you have to upload all these things. >> Yes. Suppose that can I show you one example? One more example please. >> Yeah. Okay. >> Uh so suppose what I meant by confidence. So this is Akar. I didn't show you this. So suppose I uh I this is unstructured sales registered. I have shown you suppose I upload it here in the AAR it will run I upload and classify. So it's all automatically converted into text to columns. Here it gave it the confidence score and if I click on the mapping here then it will validate and give me the confidence score here then only after I can put in the tool. But suppose if the confidence score is less suppose I have kept the bank statement intentionally that way it's confidence score is low. So when I upload and classify it it will uh I have to select the bank from here and I can see the confidence score is not 99. So it's below 99 it blocked it automatically. You can see that gate is blocked. So unless I perform all this process like it give me the tick mark it will not give me option to commit to the analytical store. So only after this is analyzed it will uh uh allow me to upload the source into the tool itself otherwise it will block it right away. >> Okay. So uh from a chat uh Praasher has a has a query. How is the chain of custody maintained for evidence and how the source evidence corrected and encrypted by the FTK manager or speed? Yeah, it's uh it's uh and connected with the back end itself. Uh I am actually right now uh stuck in the back end. I have actually forgot where I stored it, but it's in the uh uh stores. I have stored it somewhere. I know. Yeah, I can. Yeah, this all are the uh back end. It's this is where No, this is >> this is not the back end. >> No, no, this is not the back end. I have uh because there are too many files I have just recently updated it. So uh I I'm bit lost in that. So sorry on that part >> but it's still there. I I am sure of that. But right now I if you give me five 10 minutes I'll have that much of time unfortunately. But yeah we understood Pab. Very nice presentation. >> Over to you Summit sir. >> So great presentation pra I don't have much of questions to ask because I'm already still I'm absorbing the presentation which you have shown. You spoke about forensic. Absolutely. Yes. Thank you sir. >> Uh so you spoke about 42 forensic tools or something like that while your >> 14 forensic tools. >> 14. >> Yes. >> So how what are these 14 forensic tools? The names which you shown. >> Yes. I'll show you these are the tools. This is it does the Benford analysis banish manipulation lab. >> So these are so consider the tools which you're saying. So one is Benford and Benford rule right? >> Yes. >> So these are 14 rules which probably we have used and we have made a tool out of it. Correct. >> It's there are 104 rules but it's distributed between 14 uh main tools. >> So how is that 104 distributed into 14? >> Uh it runs suppose it runs the Benford. So it runs the Benford on the all the parts like it also gave me the top nine contributors to Benford wise it didn't and it also gave me the Benford summary matrix here so in one tool you can see that four rules are ran here in that way >> four rules or four variables I mean when you run it on sales so will it will the rule change I'm not sure so that's why I'm asking >> uh I mean the big one source is same but after that it is creating the branch by using that data. >> Okay. And have you back tested or how do I get a comfort that these uh results which are being generated are >> I've tested recently and one last week I tested on one engagement. uh that's all the base I have structured I have made it's from that only it's a company listed company I won't name it but uh it it generated a accurate result my I mean not 99% but the data I feed was uh I checked also with the application which data it entered into the application and it uh captured it all very accurately like I gave the confidence score right so when the confidence was not good it already directly blocked it like you either modify the data or update the code. So when when you say that you have used it for a listed company then how are you maintaining the sanctity of the data with respect to the API calls >> because I have not I have one more feature I will show you you can see that open here if I select sovereign here all the a will be blocked blocked here so you can see that sutra you cannot use the sutra here only the rulebased execution will be made here not the >> but you say now the 90% there was something 90% and it was being blocked So in that context only you say now that you have used it for unlisted corporate and >> yes >> and uh because the confidence score was less so it was being banned so the question of confidence score is coming when you're doing an API call right >> no no it's it's creates a generative generate signatures you know from the data table when the data is in tabular format it creates its own like how many database can be created it with it and then it generates the signature and counts signature and match the signature with the data it have generated. >> If you can elaborate I was able to understand something >> like uh suppose five headings are there. So it will create a signature for that like for this data we this is rulebased engine not API based. So it there is a rule fitted like You're saying it's a kind of a comfort which the to the analysis is giving, right? >> Yes. Yes. >> Okay. Thank you so very much for the presentation.