A Two-Time BIG Judge on What Separates Real AI Innovation from Buzzword Bingo | Sasibhushan Rao Chanthati

Russ Fordyce (00:00): Welcome back to the winner circle, everybody. I am here today with Sasi. He is one of our one of our all-star judges. And he's not just an all-star judge in 2025. He was an all-star judge in 2024 too. And on his way for 2026. Sasi, welcome to the winner circle and thank you for all your help with all these awards. Sasibhushan Rao Chanthati (00:00): Thank you so much, Russ. Thank you so much for the invitation. And thank you so much for all started contribution. What I was doing was recognized by Business Intelligence Group. So that is one of the things which I appreciate you. Thank you so much. Russ Fordyce (00:00): Well, recognition goes both ways, right? For winners and for judges, or for nominees and for judges. It's that's how this model works, you know. It's you guys give a little and they get a little and so we keep everybody kind of happy and value for value. But you're you've got an interesting role in this new kind of IT driven world that we're in and in AI world. Why don't you give everybody just kind of a kind of an elevator pitch on you and kind of what you're up to and your role? Sasibhushan Rao Chanthati (00:01): Yeah, so previously I was working for Tiro Price as a senior software engineer and I recently moved to Hirecase where I'm operating the same role, senior software engineer with little bit advanced responsibilities more into principal engineering services where I'm handling the architecture and as well as the development both things at Hirecase. So most projects were related to property management incident like ITSM, ITOM and SECOPS and HR HD related work which is totally compressed on a platform called ServiceNow where ServiceNow is handling all this work. Previously I was also into the ServiceNow development role at Tiro Price. The focus was more on to equity, fixed income, cash derivatives. portfolio management and also the asset management, financial asset management where it is like both private and public assets of the organization. So the same work I'm handling with multiple technologies, multiple system integration, both inbound outbound integrations and different architecture and automation process delivering a valuable product to the enterprises. Russ Fordyce (00:02): Yeah. It's a it's an interesting business to be in. I you know, you you mentioned ServiceNow. I think that's one of those brands that most people have no idea what they do and they're this giant organization that, you know, probably almost all of the Fortune five hundred use. Or or geez, I I'm sure we could look it up, but it is it's you know, it's that it's another SAP or sales force that, you know, is just in every organization. But now you've judged kind of these IT tech nominations for big and Sasibhushan Rao Chanthati (00:02): Yeah. Russ Fordyce (00:02): for Globe and Stevie for a couple of years now. What drew you into judging in the first place? Sasibhushan Rao Chanthati (00:03): So judging means like every time the scenario changes right so whenever they nominate for the awards the organization mostly into the C-level executives will demonstrate like what is the current capability they are focusing on because the organizations like IBM or whatever companies the top companies which are participating in these awards. Don't focus on entire organization or revenue. They will focus on latest capability, what they are developing and how they are using it for the organization and what are the capabilities of that capability. For example, we can take ServiceNow. So, AgentiK is one of the capabilities they have introduced and those capabilities work based on the large language model, which is directly developed by ServiceNow. And those large language model need to get synchronized with whatever data which the company is using. Of course, it is secured. So those capability they will be advertising. So totally depends on what capability the organization has produced for winning the award and drilling down into the capability. What exactly is the capability? How it will be? useful for the enterprise, for the growth and new kind of things. So that's how it totally depends on the activity, judging activity. Russ Fordyce (00:04): Well, y you know what's interesting, you mentioned the agentic push, and this there is this huge agentic AI, you know, buzzword now. And and you know, I've started playing with it on in my projects. it's it's very it's very interesting. I think Gartner predicted that like forty percent of applications would have these kind of agentic agents in them by the end of the year. I I I feel like that's really low. What's the difference between kind of a task specific agent and a chat bot bot style AI that most people picture? Sasibhushan Rao Chanthati (00:05): So agentic is totally different for different scope. For example, agentic assisted can be used for ITSM for internal process. For example, we have like thousands of incidents which are created today or which need to be completed by today and of the today or whatever it may be the agreement or the time stamp what we are using here. So that would be totally different than a customer support automation chatbot. For example, a customer support automation chatbot may reply, first of all, it will take the inputs based on a unique ID. First of all, the chatbot should know who is chatting with the chatbot. For example, I am Sashi, and I want to chat it. So it will find out a unique ID, first of all, like. For example, my email ID is common for me only. It's a unique part. I don't have any duplicate emails, right? So it will take the email ID and it will just go through the entire applications or forms or whatever, or data on my email ID. It will just go and it will review it. And after that, will give you... some kind of answers to the questions what I'm typing to the chart bot. For example, I need so and so plan to activate, or I do have a billing issue, something like that. And immediately chart bot, what it will do is it will drive to an incident page and it will ask like create an incident so that someone from the executive team or from the concerned department will reach out to you through email or phone, whatever it may be, and they will solve the issue. So that is the old traditional chart bot process which was going before. But currently, chart bot is something which is operated on different models. For example, can take large language model or small language model, which they have developed it by the organization or our organization might have developed it. Sasibhushan Rao Chanthati (00:07): they may have purchased it from somewhere and they take that element the synthetic data which is already available relevant to that customer support not not relevant to everything synthetic data available everywhere and those synthetic data will be trained with different methods like graph they can use graph graph based training or they can use like vector embedding where it will it will retrieve the information and it will Vector embedding is something like a very huge concept. If you talk like it will go for hours. So vector based databases, they will use and they will train it. So what is the advantage of this modern HHDKIs? It won't solve the, it will try to solve the problem in the chart itself because it can go and it can get the latest information of the customer from the databases, which is stored by a unique ID. and it will retrieve it, will argument it properly and it will respond back to the customer. So that is the argumentation process and which happens in different ways by using different technologies and methods. But this is the process where the customer can get the solution easily instead of going and raising an incident through Chadwood and getting help from an individual person. Russ Fordyce (00:08): Yeah. Russ Fordyce (00:08): Well the the agentics stuff is just so interesting to me because it's you're really, you know, you're you're putting people, you're putting intelligence out into the world, into your systems, to do stuff for you, you know, and and if you allow them to do it with kind of no permissions or with it with, you know, no authorizations that need it, it they'll just kinda go and act. I've had a lot of success with it, you know, fixing problems, building software, that type of stuff. You you mentioned kind of these small language models. I th I d in the conversations I've had with AI founders and CISOs and and others, this seems to be a really big trend. And if you think about it, it makes all the sense in the world. You know, these these kind of frontier LLMs are seventy billion or two hundred billion parameters. And to process all that takes, you know, all these data centers, right, that they're building and building and building. But you can get like a small language model that's really kind of fine tuned kind of to your own organization, your own data that's, you know, ten times smaller, like seven billion parameters. Does that match your kind of own expertise in building these production systems? Are you relying on these kind of smaller, kind of more fine tuned models? Sasibhushan Rao Chanthati (00:10): So there are two ways, totally depends upon the organization data and how they want to build it. For example, many companies, for example, let us take the example of ServiceNow. So ServiceNow is just providing that large language model and maybe small language models. So the LLM which is available in the ServiceNow is directly a capability now. users can go who are implementing service now for their work. They can go and directly activate that capability by activating the plugins in the administration part of ServiceNow, ServiceNow Administering. And once the plugin is activated, they can directly customize that LLM for whatever work they're required actually. So this is one way. And developing LLM is like a big thing because LLMs are already trained nowadays. They are trained with transformation neural networks and everything and they are generating response with the synthetic data. Whatever data the company is required. For example, the company may need like financial data or maybe some other data. So maybe some company may need quite retail related data. So whatever the synthetic data is, those data they are training the small language model with those data, okay? And they are directly giving it. The only thing is we have to train that it should be a secure. We need to implement some security standard like secops and everything. And once that is implemented, because the data should not go out because so once implemented, if we want continuous prompts and a little bit of coding activities also required based on the programming language what the small language model is developed and we will train them. Russ Fordyce (00:12): It's a Russ Fordyce (00:12): Yeah, it's interesting that that so go ahead. Sasibhushan Rao Chanthati (00:12): We can train them so that it can synchronize with our business process. But the only thing is like developing a model, a small language model for a huge organization is required because they own their own products. For small organization, the services, whatever service now, Amazon Web Services, Microsoft, Cloud, Azure and everything, they are providing the LLMs to the customers. where they are training them with their own data in a secure way. Russ Fordyce (00:12): Yeah. Now I I I've also heard of a lot of of engineers and and programmers actually moving towards kind of a hybrid model where they'll have, you know, a specialized, you know, small language model that's runs, you know, ninety percent of or eighty percent of their queries and then off to kind of frontier models. Are you seeing any of that kind of programming work going done or kind of it fails over to the LLMs? Sasibhushan Rao Chanthati (00:13): No, that is actually the hundred percent many organizations are focusing on so Large language model is nothing but it will have the entire data of universe the world when you ask any question I want to travel to Paris from United States from which location it will of so those are all not required for the Organization right so we don't want to waste our Data providing all those information, right? So obviously, it totally depends on the domain specific and relevant domains with support to that organizations. So that's why this situation will be handled. Because let's take like open. Russ Fordyce (00:14): Yeah. Russ Fordyce (00:14): Well that we've Go ahead. Sasibhushan Rao Chanthati (00:14): I'm sorry, ahead, ahead, you go ahead. Russ Fordyce (00:14): No, it just we've seen we've seen this kind of, you know, this all the news reports of, you know, token maxing and you know, companies getting surprised by bills. And so these, you know, these things like ServiceNows enables you to kind of plug in your corporate API into Anthropic or OpenAI, you know, the bills start ringing up. And now we have these kind of Chinese AI providers that are leaning on kind of smaller, cheaper models where they're coming in and undercutting prices. you know, radically, what does that shift kind of signal about where we are and where the real kind of competit competitive pressure's heading in the in the AI landscape? Sasibhushan Rao Chanthati (00:14): So the only thing is like totally depends on the company standard what company is providing that AI model like for example, if it is like Amazon or it's a trustable organization, right? So and those organization products are also cheaper. It's not like any other companies or any other country products which are very cheaper than the current standard organizations which are available in market like Microsoft, Appian. Appian is one of the SaaS related platform oriented organization which recently coming into the market. And also there are a few organizations which maintain the standards. And also they also have the secure process rather than maintaining the standards. Security is also important and it plays a different layer security, not just a single layer security or whatever maybe. And it's a big concept actually. It's a big subject where they implement the entire secure process. Only standard companies can able to handle that secure process to avoid data leakage and everything. And those APIs, the APIs what they develop. set of code or maybe it's a it can the IPA can be used for integration process. Those are also secure. They don't even have any leakage or anyone can interrupt in the API and get the data. So that security is important. One of the things and data, whatever data the company is having like customer data, their financial data, their sales data. Everything is also important, right? It's not like that. It's not like the small companies or whatever unsecured organization can take the sales data of one company and give it to others. So these all things play a very important role when considering the cybersecurity and also the data security. So only standard companies will provide that standards and also they will give assurance also. It's not like they told us that Sasibhushan Rao Chanthati (00:17): They have mentioned that we will be providing the security standards, but they will give assurance like if something happens that would be their responsibility also and Second thing I don't think like some other small company products or maybe come products coming from different countries are cheaper It totally depends on the cost of labor Who is operating in that organization totally the cost will be that and the product price may be cheaper you compare with AWS or any other standard organizations. then according to me, I don't think like it would be a good difference. So you should suggest me like you should tell me like. Russ Fordyce (00:18): Yeah, I Russ Fordyce (00:18): I I've I've heard they're not as accurate, they're not as responsive, there's not as feature rich as well. And so it's you know, you're paying you're paying less but for a you know, a less capable product. Sasibhushan Rao Chanthati (00:18): Right. Russ Fordyce (00:18): moving on to like customer support automation. You were talking about service now and you know, support automation seems to be kind of one of the biggest sectors that that AI is is having an impact on. My own my fiance actually, you know, had to make a call into a a medical office today. And you know, she called once and got a human being and she called back, got disconnected or something. And the second time she called, she got an AI. And she was a little put back, she's not a very technical person. But she was able to get through it and it got through it just fine. And she realized it was AI. And I said, and you know, in a year and a half, you you probably won't even recognize that that's AI and it'll just work. And I think Gartner's predicting that like 20 or 30% of service agents will be replaced with generative AI this year. So this impact is going to be huge on this sector. I think they said it like, you know, 100% or 80% of of the routine kind of transactions on support. will actually be handled by AI relatively soon. Does that kind of jive with what you're seeing in this new kind of automation realm and the support angle? Sasibhushan Rao Chanthati (00:19): So support may be like customer support will be like through many ways. One, the customer calling to the calling the organization and asking for the support and second through chat boards and third through email sending an email to end customer support asking for so and so help or something like that. So these all things are getting synchronized with one system so that one should understand like what's going on with that customer exactly. So those things are handled through ITSM process like incident management or and that incident goes to change management and again if the customer is completely complaining that issue again, again it will be a problem management or something like that. But those all things are now handled with one system. where they are synchronizing everything. It's not like three different process, three different departments or something like that. And when someone answers, when someone calls the organization for help regarding their customer, they have their purchase, their product or everything. All those telephone calls, emails of the customer and whatever they are sending, right? It's not like one person will only call the company. They can send an email, right? So all those are getting synchronized with one system with that customer, entire customer record. So answering the call is an old technology like whatever the customer is asking, the robot will listen to that and it will translate language and it will take a proper response and it will deliver. So those are like... before it used to be specific questions and answers. Now it is not like that. They are thinking like human beings. The reason behind is like those are getting trained, those language models are getting trained a lot on that. And they are delivering the appropriate answers. they are training every day, like whatever situations of the customers are there. Sasibhushan Rao Chanthati (00:21): all the data and everything and they are retrieving data related to customers and they are getting trained every day like every hour also, so the organization. So 90 % accurate report will be delivered by the telephone call when we are talking to an bot. Russ Fordyce (00:22): Now you've been in this market for a while. You've seen kind of a lot of these kind of trends come and go. We've, you know, had this digital transformation. We went to the cloud. You know, we've IP enabled everything, everything's online. You but you've been really close to it both on the customer side, you know, working on these projects and also reviewing all these nominations. in your perspective, what's the biggest shift other than AI kind of that you've seen in the last year or two? You know, both from your customer perspective and both in the nominations. Any any big kind of call outs that you're seeing, any big trends? Sasibhushan Rao Chanthati (00:22): Yeah, the big transaction I have seen is cloud, the legacy systems, most of the systems like the databases, which got migrated to cloud, mainly AWS, Microsoft Azure Cloud and Google Cloud. So that migration projects are very huge. Those are the migration projects. It's nothing but they have to remove their own... database whatever it may be the server windows or Linux server and they have to move the entire thing to the cloud which is hold by some other company and lot of services have shown up like storage services and pay to use like how much you want to use you have to get pay only that amount of money it's not like you need to purchase the entire server and even though if you have space, you can leave it alone. So those are some of the main advantage concepts which will happen in 2018, 2019 before COVID and at the time of COVID, the migration projects to the cloud. And companies also saved a lot of money instead of maintaining the entire data center and everything. They can rent whatever space they want from cloud. So that is one of the major change happened in the organization. Most of the projects were cloud migration projects. And they are decommissioning the traditional databases and everything and going to cloud. And after that, I'm seeing like this AI trend where LLMs got introduced and small language models as well only synchronized with the companies traditional data only not with all other data. So and outputs like AI outputs are also validated and they're giving correct responses also before production issue. I mean before moving that to the production to the live or whatever it may be. So the validation process is also a process where we have to do right. So that outputs are also very accurate very accurate and Sasibhushan Rao Chanthati (00:25): the testing processes, automation processes and everything. So this is one of the trends I am seeing. These are the two major trends. Russ Fordyce (00:25): Yeah. Russ Fordyce (00:25): I would think the role of FinOps is also, you know, really getting more important as as the you know, as as all of us kind of tap into the power of AI. Are you seeing companies spending more money or time on FinOps? Sasibhushan Rao Chanthati (00:25): So financial operation is nothing but it's a cloud financial management discipline, It totally depends on companies, what they use. The SaaS products mainly support these financial operations. many companies which have good revenue, they will adopt multiple SaaS products, like software as a product, like Appian, ServiceNow, whatever it may be. But a few companies will synchronize all their SaaS products into one single SaaS product, where they purchase only one single SaaS product. So that's the reason companies may spend money or may not spend money on building their own SaaS products. Because financial operations mainly go through workflow process, or we can call it as a business process management. It goes step by step. So as per my knowledge and I have seen like working lot of with work, working like judging the companies like I have seen like hundreds of companies where they describe everything, what they are doing and everything. So development of the products, I haven't seen any company investing money to develop their own source product for the financial operation. But they purchase that software as a service products. and they will utilize it or they may customize it as per their business requirement. Russ Fordyce (00:27): I wanna Yeah. I wanna turn to the softer t side of IT with you. You did a lot of research on IT burnout and the kind of human cost of this always on engineering culture. And it was done you you did it, I think, back in twenty twenty four, right? You started that? Sasibhushan Rao Chanthati (00:27): Yeah, it was like like 2021. yeah, 2020 or 2021. I got the copyrights of it from US Copyrights Office. And once I got the copyrights of his permission, then I have implemented it in three, four phases. Actually, it's like the traditional what I was doing is I was building a chart board to handle that burnout mechanism and everything. So after that, like Russ Fordyce (00:27): Is that right? Sasibhushan Rao Chanthati (00:27): vector embedding came up, like I was using MongoDB and developing vector embedding. And the results I was taking out with dummy data. And after that advanced vector embedding process I have done, and after that I was integrating that entire chat, what I removed, vector embedding I removed, and I just integrated with an existing traditional open source LLM. And it is giving wonderful outputs. after training that. So the cost and time is when I was doing a chat board was too high at that time. Like it took a lot of time for me to develop the chat board and test it. But the latest one is like time saving part, exactly speaking. As I am using open source, so I don't know about the cost, how much the companies charge. But actually, truly speaking, like it's a cost saving. Like I have saved a lot of time. Russ Fordyce (00:28): Yeah. Well what I liked is in reading about it was you really you're you're using kind of AI to detect, you know, burnout through kind of how people work. What was the moment that convinced you that this was worth building? Sasibhushan Rao Chanthati (00:29): So burnout mechanism is, so when I was, I got an idea because I was working like for healthcare domains and everything. So I understood like what healthcare domains mostly focus on rather than financial domain, I do have some exposure on it because when I was doing my masters, I was doing some internal healthcare projects at universities. So one of the challenging part where even Howard and MIT cannot able to solve is burnout mechanism. because burnout mechanism can be handled in many ways. And as per an IT specialist, I just want to check it out how that can be handled through software as a service or maybe you can build a platform and how that can be handled. So the only thing is like I have taken only an organization where whenever there is an employment related activity going on within one employee with or her manager or with peers and everything they directly go and talk to each other or they will exchange emails texts or chats and everything so if they are going and talking to each other we don't have any control because we can't record that but when they are chatting or when they were using like symphony or teams The systems can identify the actual language what they are using and are they are stressed or are they are having any issue or they are struggling to get that work done or they are waiting for approvals, permissions and everything. So those are the keywords I have taken from the exchanging of chats, emails, calls and everything. And I have done like some a huge vector embedding process. One is like taking the data from the chats, chat activities based on an employee unique ID, which may be employee ID, which is unique. And at the end of the year, it will finalize like, and it will also throw reports to human resource generalism. So these are the conflicts going on and everything. Otherwise, human resource generalist or whatever Sasibhushan Rao Chanthati (00:31): the department they have, they don't understand what's going on. And also I have added stress, one of the things like working overtime. Some people work overtime to make money that does not have any impact with them or to the organization. Some people are forced to work overtime because they don't have resources. So those things also we can pull out by their exchanging of messages with. using teams or whatever it may be. Because I can show you like I have worked eight hours, more than eight hours, I cannot handle tomorrow or something like that, then the system will understand what's going on. And it will show an alert to the concerned department saying that so here the conversation between his and the manager is going on like this. and other person from the generalism department can involve and settle that process. Russ Fordyce (00:32): Well, it it's interesting to me. The the the role AI is playing, I think is interesting to this kind of burnout effect of especially in kind of the roles you're talking about, is AI is supposed to be this enabler that's giving you more productivity. and I found that it it gives you more to do. You know, it it tends to make us busier because now we have new skills and new things that we couldn't do before. how are you finding it, you know, impacting your your You know, that that sentiment that you're getting out of the tool and and the research that you're doing is AI making things better or worse? Sasibhushan Rao Chanthati (00:33): Yeah, is before like in the first module when I released the first paper the monitoring usage was not there. It never used to monitor. We have to view that inputs and it has to calculate. Now in the second vector embedding model, it has done some analysis by using vector embedding, semantics, search and everything. No analysis monitoring, forget about it. I don't even know about the cost of implementation because I am using open source. identifying and everything. Now it is measuring the results like it is totally measuring the result and it is repeating the activity again and again. So this is the advanced stage I have learned but I don't know like I have found out like my findings were on measurable results and repeated activities which are going on. But the only thing is like I can't know I don't know like what will be the future what steps will be useful for the future and what will be the new thing which can we implement in future? So that is the thing I'm into more research here to find out like what would be the need for or want of the people or the customers who need this project to be an extraordinary capability like upcoming capability or something like that. Russ Fordyce (00:34): Right. Well Sasi think Sasibhushan Rao Chanthati (00:34): significantly it has done lot of things. Russ Fordyce (00:34): It has done a lot of what? Sasibhushan Rao Chanthati (00:34): Significantly, there were a lot of improvements in the last three, four years. The new capabilities are getting introduced, and most of the things are getting synchronized with the new capabilities. So finally, I want to check what would be the next capability which users want. And if you know that, let me know. We'll do some research on it. Russ Fordyce (00:35): We'll start a new company and give them exactly what they want. It's it's it's it's moving at such a pace now. AI is enabling, you know, such a pace of change now that it's really it's hard to keep up a lot of times. you know, even the software we're using to record this, it seems like, you know, this is getting a step improvement. It's already a good product. It's getting a step improvement every week or two weeks or month. they're doing a whole nother release with new features, new functions, new layouts. And you can now with AI, and that's gotta be Sasibhushan Rao Chanthati (00:35): That's right. Russ Fordyce (00:35): exciting as a engineer and also you know, on the burnout side, a little bit disheartening too, that you you now have all these new capabilities that you can deal with. Sasibhushan Rao Chanthati (00:35): Right. So can you elaborate more like what exactly that you are focusing on? Russ Fordyce (00:35): Well, I mean it like as you see it, you know, y you now have these new capabilities too. We have this agentic AI world, these chat bots, you know, chat GPT that can go off and do things for us. And as an employee, as a business owner, I find it great because I have all these new capabilities, but I also find it that my brain now is almost overstimulated all the time because now I'm not just, you know, the I'm not just doing the operations of this business, I'm now doing coding and I'm doing you know, more marketing and product development than ever. And so it's just it it seems to be that, you know, the burnout factor might get a little bit higher before it gets a little bit lower with AI. Sasibhushan Rao Chanthati (00:36): Right, that's what right we have to identify the correct product or whatever it may be like if you want to develop a product by ourself or if you want to purchase a product from other organization and train it or develop it by ourself once the purchase is done we have to identify the exact product which is required because there will be a lot of AI capabilities in the market and some may not be useful for you for any organization right they purchase it and they keep it that won't be that won't be any useful thing for them. So too many products are there I agree with you but you said like chat GPT and everything right so they can suggest you or people can suggest you or when you browse on when you browse on internet like you can also get a right product for the right business capability what you want to handle through automation. So that research we have to do otherwise it will be a huge loss, time waste and everything. Russ Fordyce (00:37): Well, Sasi, thank you so much for all your help with all the scoring and all the awards and all the success you're having with your jobs and your research. I really appreciate all the work that you've you've given us. So thank you very much. Sasibhushan Rao Chanthati (00:37): Thank you so much, Russ, and have a good afternoon. Bye. Russ Fordyce (00:37): You too.