In this interview with Manthan Trivedi, CEO of ManCorp Innovation Labs, we talk about the problems that the law industry in India is facing and how ManCorp Innovation Labs have helped High Courts in India like the Jharkhand High Court, Patna High Court, and Supreme High Court solve large-scale problems with innovative and AI-based systems.

Interview highlights

  • Large-scale problems in justice administration in India: systemic challenges, infrastructural and digitization issues.

  • The importance of digitization to unlock productivity and efficiency.

  • Overview of the tech systems/portals implemented in Jharkhand High Court, Patna High Court, and Supreme High Court. What do these systems make possible, that wasn’t possible before?

  • Artificial intelligence, and differentiating facts and opinions by machines

  • The system implemented in Patna High Court, how does it work and what sort of data is being observed to be able to predict

  • What is SUPACE, how it works, and where it is implemented?

  • Why is it essential to have a unified system for all the courts?

  • Where is ManCorp Innovation Labs going in the next 5 years?


Manthan: The pleasure is mine. Thank you!

Namit:  Okay, so, so Manthan, you’re the CEO of the company, and I want to just start with understanding where you’re coming from, and in terms of the problems that you found in the law, injustice space in India before you innovated anything, what is it that you were seeing in the market that needed to be solved?

Manthan: Right. So this is the story back at the end of 2015, when I was studying at Harvard, we were writing a lot of papers, we were reading a lot of research papers, and devising data models in order to make decisions. And there came out the idea that some of these policies that have been made or some of the plans that we executed, even as a student working with the Harvard Consulting Group, you know, we’re trying to create efficiencies in the system, and what I realized is that they have been systems in place in countries such as America for over a century now, and they are just trying to improvise on those systems and trying to make the systems more efficient, whereas, in countries like India, Bangladesh, many African countries, these developing nations, we lack necessary systems in order to operate or to make informed decisions. 

So, a major gap was to be bridged between the collection of data, what kind of data to be collected, what can be done with the data that is collected? What do we want to be done with the load of unstructured data that lies within various systems? 

So, the initial idea was just to come down to India to understand So, to give you a little more further background about myself, I’ve never lived, this is my first time living in India, or doing anything in India, for that matter. I haven’t studied in India. I haven’t done my education or anything of that sort. So my first goal was to come down to India to understand how business happens here. Why? Because I see it as a plain economic choice to make a data-driven decision making and why does it not make economic sense in India for that matter, and if it doesn’t, then what makes sense? So, this was my first approach. And then I understood that there are a lot of lags in the understanding of what data really means, and what one can do with it. And this was, I mean, there were many reasons contributing to this but primarily, there was one of the reasons was that you know, there is no hardcore research that really happens in India, whether you talk about deep technology, whether you talk about sustainability or any other theoretical fields, you know, there is a research or there is a revolution, as we talk about that happens in India, which is sponsored and like say there is something been proven in other country and that has to be replicated in the Indian context, that is the kind of researches and studies that happen with you, but if you know, we are to explore a new territory altogether. Suddenly the entire market or everyone sort of becomes uncomfortable with this idea. 

So, primarily, this is where we began. Luckily, soon after, we incorporated the company, we incorporated the company with the idea that you know, we will try to bring in a solution where we can sort of channelize this unstructured data When I say unstructured data basically, I mean that, you know, paperwork, wise voice communication, any kind of communication that happens between a human and another human. 

There is a lot of unstructured data that is created on day to day basis. Just to give you an idea about what happens in the judiciary. Some studies suggest that in the Indian judiciary every year, 11 billion individual sheets of paper is brought into the system. Now, this is a quantum of unstructured data, that is nearly impossible, if not actually impossible for the entitlement system to deal with it, you know, as we go about our day to day business, but there are already technologies that solve the problem of structured data, like when you already have a CRM system, a SAS system, you have already fields inputted into that, and then you can basically try to channelize that and understand, you know, where you are, and what you have, and which is what primary the work has been in India, whether you talk about the national judicial data grid, which again, tries to capture, you know, aspects of these metadata fields, what type of cases when it was instituted, how old is it, whether it’s spending, whether, you know, how many are disposed. 

So, this kind of data is being worked around with. But there was a serious gap in actually trying to address the problem of unstructured data. So, actually, the files that you get the paperwork that you get, if you could actually make sense out of it, if we can actually channelize it.

So, one area of work that I saw that was happening in India was research. So, there are a lot of leading research portals in India, which are doing this work of unstructured data, classification, and categorization for the purpose of channelizing and further So, some of these companies are basically taking all of these judgments and acts and presenting it to, or making it available to, you know, the legal fraternity for ease of access, and trying to implement AI technologies, where it can be easier to even access those files even faster, or even find more relevant judgments or acts that are relevant to your case. But beyond this, there wasn’t much else happening.

 Another trend that I observed was that there was an inherent resistance towards process automation. So, it was I mean, it in my initial talks with several people across various industries, one of the first things that we got is that, you know, it is going to give rise to unemployment. And then one of the first questions that came to my mind was that, you know, if process automation leads to unemployment, or for that matter, any economic exponential productivity leads to unemployment, then why would anyone agree to it? So, one of the examples that came to my mind was, in the initial days when computers were introduced to the world, right, so that they are human computers, actual people who are employed to say day in and day out and actually calculate large sums of numbers, and they will call computers and then they will replace with the digital computers that we have today. 

Computers haven’t gone: The computers had become then accountants soon after this right. So, their job role became more specific to what those computations were supposed to be done for instead of actually doing the clerical job of computing right. So, that is an example of process automation that I found was missing here that you know, it was directly seen as AI and process automation directly taking away jobs, but that is not really the case. So, if you actually look at design thinking, then you can actually design an industrial output and still, eliminate certain risk factors. So, as a design, if we actually try to eliminate processes, that are not going to cut down jobs, but are going to still bring out economic benefits, human resource benefits skill development benefits, then that is something that is valuable to everyone because, at the end of the day, no one likes to do the tedious jobs that are repetitive in nature, and that is something that you’d rather spend time doing a qualitative job where you’re actually applying your mind and that is the feedback that we got from everyone. So, there wasn’t a problem about process automation or AI as such, but there was a gap in understanding that was the primary area that we sort of encountered as soon as we entered in this fraternity.

Namit: Okay, I have two follow-up questions and one is why is the structuring of data or why is digitization, so, essential to be able to unlock productivity and efficiency in other aspects, why is this so fundamental? 

Manthan: This is the same.  I mean, see, the idea is manual labour, whether the manual labour is physical in nature or is mental in nature, it has a certain amount of limitation, right. 

Say there are 24 hours in a day, there is a limitation to how much time can be spent at work by a human there is a limitation to how much work can be done by a human in a given particular timeframe, right. And if you actually mechanize this process, which is what happened in the post-Industrial Revolution, where they actually brought in types of machinery to do the things that humans were doing, which started with the textile industry first right. So, what happens is that machines can keep on doing the same job repetitively without or not necessarily without, but essentially reducing the number of errors essentially working endlessly, right. You can, we can let the machine work 24 hours and it wouldn’t get tired. So, that was the force crash funnel of why mechanization came into the picture. 

And now mechanization in the physical space happened at a very early age. And soon after that, now, we are looking for, now we are at a phase where many production lines are automated, many of the design aspects are automated many of the even if you look at construction industry or any other industry, there are new machines, new equipment that are coming in every day or every year for that matter. So, these are efforts in order to try to make a more economic sense of the activities that you’re conducting. Now, it is that we are delving into the manual labour that is not physical in nature, but in mental in nature. Now, when you’re talking about manual labour, you know, you’re talking about all the work that you do of reading, reviewing writing, talking and everything is where you need to apply your mind just as you do in your physical work but at the same time, you know, if you need to automate this process or if you need to bring mechanization to this, then you need to basically be able to provide this data this information because, at the end of the day, it is not, I mean there are systems which in fact in AI how it started was with pre-programmed systems right which we should call software today. Just regular programs that would be based on a certain input they would compute a certain output. But increasingly, as data increased in the world and as the field of AI research, increased, and progress, what we have come to is the ability to be able to automate and to be able to reduce the tasks that are not necessarily fixed in nature. 

So when I say fixed is like, say, you know, filling a form is a fixed job. Now that you know you have the name field, you have the address field, you have the phone number field, you just need to fill those items. So if I have the data that I can automatically fill, that is simple programming. But, you know, when we talk about summarising a case, for example, then, although the points of summarization may be common, for example, you know, you might call out reasons of the judgment, the cases the case-law cited, the arguments presented or the issues presented and the reasons have a court on each of those issues. Now, these might be the components that are repetitive and common in nature for all my summaries but it is not common as to where this information lies within these documents, right. And here’s where big data comes into the picture, where big data is able to analyze patterns and is able to deduce, with a certain amount of accuracy, that this is probably what the summary should look like. And this is probably where this information lies and is able to pre-process that.

So currently, we’re at a stage where  AI is able to do all these tasks but still requires human supervision. So the element of human employability or the automation of the complete human mind is still a far-fetched idea. 

Namit: I understand. So with this backdrop, what has been, let’s just get to the product and the innovation. The systems that have been implemented in Jharkhand High Court, Patna High Court, and the Supreme Court, what are those systems? If you can tell me from a practical user point of view? Who is the user? And what are we able to do now that we weren’t able to do before?

Manthan: So the system that we implemented in Jharkhand is a very interesting system because that is what basically ended up being developed or evolved into space eventually. But the problem statement was very simple. If they needed a smart code solution, targeted towards the section, these are two cases. Right? Now, the problem is that there is there’s a dearth of people who needed to hear the matters. Basically, the judges, number of judges who are actually required to be in the boat, and the number of judges who are actually employed, you know, that there is a shock, there’s a big gap between that. And plus, in a state like Jharkhand, you have a lot of criminal cases, especially section 302, and they wanted us the mechanism or using new technologies, where they could expedite this process. 

Another problem was also that, you know, many judges are transferred. And then what happens when they’re transferred is that a judge would have heard this matter, and on the next day of hearing another judge hearing this matter, and what happens there is a gap and becomes a latency between what the parties are trying to argue, and the what judge understands about the case. And, you know, then, obviously, the time taken to decide on that extends because of these factors. So, we devised two pieces of technology. One is optical character recognition, which basically also already a process of digitization across the country is happening, even in all High Courts, is happening. 

So, we basically took this digitized, I mean, the scanned PDFs, of these documents, and converted them to computer-readable text. When I say computer-readable text, they are basically 1s and 0s, so that the computer can decipher patterns from within them, and then we can do Intelligent Automation if needed. The second step was that one of the big concerns is that we don’t want to automate decision-making odd, we don’t want to give sort of an opinion to the judge, that would nudge the judge in one direction. 

So one of the biggest powers of big data is prediction algorithm, it’s very easy to predict a certain thing and be able to say that bit a certain amount of probability, this is where it is heading towards. But this is something that was not the need of the system because this is something that would create bias. Right? Because just because it is more, I mean, not that that’s a different discussion altogether. That’s a judicial space and, you know, a discussion like that. But so we devised this system. could basically aid in decision making without opinionating. So we devise of fact, an extraction chatbot, which will be basically able to read through the entire document at the speed of nearly 10 million characters per second and be able to decipher any information that a judge requires. 

Now, the problem here was that you cannot just create the system to decipher any information, just like you cannot tell any human to decipher any information, you have to write the human in order to understand what that information is, like, say I understand what an issue is, but I’d know I may not understand what a legal issue is. So similarly, we have to train the system for these things. So we basically sitting with the judges of Jharkhand High Court, we created an exhaustive list of about 120 to 160 questions so when what every judge requires, while hearing 302 matters should be facts only. And more importantly, when the system would give answers to these questions, it would also highlight within the document where it has fetched this information from. So at any point in time judge is not driven into bias on is not nudged, to depend or rely on that information. Although their information may be correct, or maybe incorrect, the judge will always see where the information is coming from, and whether he agrees to the output of AI, or relies on his own intelligence to, you know, replace the AI-generated output with his own output,

Namit: The boundary between fact and opinion, I think it’s so diffused, that maybe for not just for you, but other players also in this, in this segment isn’t in this market? Would this be a difficult challenge to scale? Because even if you have been able to give, let’s say, pre-programmed a certain amount of questions, and telling, basically, with the instruction that this look, this is what the fact looks like, this is what the fact doesn’t look like. But would there be some sort of dependence on for example, where the information is even coming from if it’s being picked up from parties briefs, if it’s being picked up from previous judgments, there would still be these challenges that will exist?

Manthan: Precisely. So, this is what we had identified in the early days of implementing this system and pilots are doing many pilot tests in where various judges would have different opinions or would have a different perspective as to what or rather where the information is to be fetched from what can be deduced or can be classified as a factor versus what cannot be classified as a fact. So, taking this problem into perspective, that is what we eventually evolved the system to basically not just learn about the legal context and what a legal fact means or a legal issue means, but be able to also replicate human activity at a user level. So, we basically created this sort of an activity monitor for the sake of simplicity, an activity monitor where you could basically see, you know, where you as a user while reading a case, you know, what parts have you highlighted? What parts have you called out? What have you used in your making of brief, or what kind of research are you using, and whatever processes that you have done repetitively. And that you see that you agree that this is where I want the information coming out of, then you can add a click of a button automate that particular process, we are trying to reduce designer bias by pre-training the system and giving it to the user. Instead, what we are doing is giving the user an AI-powered framework where whatever work that they do, they have the power to automate or not automate those tasks.

Namit:  I see what you’re saying Manthan. So if I get this right, first of all, the user, when you talk about you, the ‘you’ here is the judge, right?

Manthan: In the case of Jharkhand High Court, yes, it is a judge. In the case of the Supreme Court of India, SUPACE,  it is primarily the law researchers or law clerks who assist the judges.

Namit: Okay, so those would be the extensions of the judge in doing similar activities?

Manthan: Correct.

Namit: Okay, so, so my question was basic, actually, I just wanted to summarise what I think you’re saying and just to make sure that I understand right.  So that in order to increase efficiency, there’s one way which is to train your language models to understand what fact is. what fact is not, and what you are training at that point is only the result of the activity that this is ultimately what it ended up resulting in. And now your language model absorbs that. And the other way to do that is not just to be acquainted with the result, but also the process through which the result was obtained. So I see and what special advantages does that end up giving us over and above? For example, just observing what is a fact? And what is not a fact?

Manthan: It opens a world of new opportunities, right? I mean, we can end up discussing what is a phone for five hours? I mean, you’re talking about essentially, what is the importance of data, because today also, there is a process there is a method to the madness, right? 

There is no one who just randomly wakes up in the morning and passes a judgment or makes an opinion about something there is a method to madness. What we are essentially trying to do is trying to bring in for a lack of a better word, a simulator of this process, where you can simulate beforehand. You can actually observe what is it that you’ve been doing, you can actually redirect yourself or other you can make more informed decisions, you can be more consistent with the line of thought that you’ve been having in the past. compared to today, and there are endless benefits to what is the benefit? I mean, what uses is data for right. So there are endless, and there’s an endless potential today. I can just sort of give you opinions as to what are the potential outcomes but unless it actually lives up to this potential, or you know, actually had this practice, we don’t know, right?  Like, say when Facebook started. I’m not sure if you’ve seen this documentary on Netflix called the social dilemma. Right? It talks about how Facebook is able to predict that someone is homosexual or heterosexual years before even they realize. So now this was not a problem statement or identification that we’re going to use data. But this ended up being a use case, or this was this ended up being that, you know, the power of data was so much that they could even deduce such facts about a person that he or she himself or herself would not know about themselves for many years until the people with data know about

Namit: True. True. So the program in Jharkhand High Court has been rolled out. Has that been operational?

Manthan: It is not operational in a full capacity where every judge listening to 302 matters is using the system. That is not the case yet. It is still being piloted, where only a handful of judges are actually testing the system and giving feedback, whether they agree with the system output or not. 

Namit: What have been the results so far? 

Manthan: So some of the suggestions that we’ve been getting our feedback that we’re getting are things that we have anyways implemented, or improved upon with space development. So we are just integrating those solutions into Jharkhand High Court. And apart from that, they’ve been wanting language models of translation and the ability for the computer to be able to communicate also in multiple languages, which we are working on right now.

Namit: I see. Okay, so let’s move to Patna High Court. And I want to understand what the problem statement there was?

Manthan: Right. So with Patna High Court, we are trying to basically see as chief justice of any court, whether it’s the Supreme Court of India, whether it’s any high court, you know, as chief justice of any court how can you optimize the work that you’re already doing, which is basically managing the allocation of cases. Right. And be able to allocate them to the right judges who have had the experience and have given good judgments on those topics on those subjects, and have proven to, you know, be effective and speedy disposal of cases. 

So looking at those historical records, we basically created a system that would suggest you know, What the allocation of cases would look like should look like to optimize time, and the Chief Justice could basically shuffle that reshuffling again and change it according to his needs. Again, the idea is the same where the AI is trying to observe and trying to replicate as much as the activity of the user, which in this case would be the chief justice trying to optimize the allocation of cases.

Namit: And what sort of data is being observed in order to get to the prediction?

Manthan: Previous allocation of cases. So that would mean subject matter. Same subject matter the benches, who have been, you know, where which cases have gone to which bench or which individual judge in the past and that will be the data to, in order to basically suggest…

Namit: …individual performance in terms of the speediness of disposal? For instance?

Manthan: Yes. So disposal? Like, yeah, basically, a decision being taken on any subject matter was an important parameter for this

Namit: All right, and, and what’s the status that hasn’t been pilot tested? Is it been implemented?

Manthan: It was,  shortly piloted It was not, it is not implemented, it is not in use as we speak right now, it is not to my knowledge. But this was a system that was developed, piloted over there. And this was an activity we conducted, also, as an exercise to sort of provide feedback to the Supreme Court AI committee, as to you know, because, as a Supreme Court AI committee, you know, tries to identify bottlenecks within the entire system, whether it’s administrative in nature, whether it’s judicial review or legal review in nature, it’s to identify various bottlenecks and how technology could solve or aid in these areas. 

So apart from process automation, in audio order, aiding judges in terms of getting briefed about the matter, this add administrative issues was another area where, you know, it could actually potentially benefit the Justice delivery system as a whole

Namit: Administrative issues such as?

Manthan: Such as which cases being allocated to which bench or when judge, based on the previous experience, let’s say some judge has great expertise on commercial disputes, versus someone has great insight in criminal matters. You know, these are things that a judge intuitively knows by experience, or by working in that particular High Court. But as you know, Chief Justices keep are always transferred, elevated to another court or to the Supreme Court of India, and what happens is that you then there is, again, a learning curve, and in that learning curve, what when a new person takes over, it actually leads to inefficiencies. And so these are the areas that we’re trying to bring stability to trying to add value in terms of whether it is the Chief Justice sitting right now or a new chief justice to come in the high court, the High Court is going to behave a certain way.

Namit: I see okay. Okay. I want to know about SUPACE and how it was started from basically the point of view of the AI community, and then when you came into the picture, what is the innovation and what is the current status?

Manthan: Right. So, after implementing phases one and two in the Jharkhand High Court the Supreme Court of India had basically constituted a committee for the use of artificial intelligence in the judiciary.  I was fortunate enough to be a member of that committee, where basically we will discuss various bottlenecks and from what we had identified we are already hitting the target right. I mean, we are already creating a system that has the natural language processing capability that has the OCR capability, that has the machine learning capability to understand user activity. 

At the same time, we are trying to eliminate biases as we speak already, by giving the user you know, the power to automate a task versus the developer of the product or train the system to automate a task. 

So these were some things that we are already we already had in place. And this was something which was highly appreciated by the AI community because they were also looking for, you know, solutions, where they could fast track a given particular matter type of case or they, could put or they could say, you know, automate or fast track a certain step in the Justice delivery system. So even, like while SUPACE was in development, we are exploring the use of this technology also in registry, for example, right, when you submit a case there is a process of identifying defects, and whether you accept the case or you reject the case with the list of defects. So now, this is, again, a repetitive process, and the defects are again, you know, they are not necessarily common, but they fall into common fields. So, then we are training the system to understand those areas as well. But primarily, SUPACE is again, trained, or being used by law researchers, law clerks working with the judges, in order to review the case files in order to make the briefings, you know, in order to do the research, so that essentially, at the end of the day when the law researchers working in the Supreme Court are interns, so when they leave the internship or a new intern comes in place, they already have a standard protocol they already have the data that you don’t get that can be shared, the intellectual capital can be grown. 

Namit: Manthan, for those watching without the context of what SUPACE is, because you’ve seen a lot of news and the news often covers that, okay, so SUPACE was rolled out, this is what SUPACE stands for. So for somebody lacking context of what exactly the product is, would you mind explaining, giving a summary of what it is?

Manthan: So SUPACE is an AI-powered, smart office solution at its core right? This is a portal where you are able to do all activities digitally, that you would otherwise do physically, which is read your paperwork, be able to highlight pin, tag information, be able to copy-paste information into your, you know, whatever document that you are drafting briefs that you’re creating or anything else that you’re drafting. So all of these activities you’re able to do on this platform is the first step of what SUPACE is.

And the second aspect of this is that you know, that there is an AI-powered chatbot, which has already been trained to understand facts, has been able to understand the legal terminologies and context of legal documents. So is able to answer contextual questions of the documents. So instead of having to read 1000 Pages document, you know, or a 500-page document, in order to identify what evidence was presented at what stage, you could directly ask a chatbot as to you know, what evidence was presented at what stage or what was found, or what was said, by a certain party, or what was argued by a certain party, or was deposed by, you know, the doctor or anyone else for that matter. So these are the kind of information that the system would be able to fetch in order to get, you know, a faster understanding of the cases. 

And the third aspect is that the system will basically give the user the data of all the tasks that they’ve conducted so that they could actually understand what a representative is, what are some repetitive tasks, what are some complex tasks, so that even without actually, even before you actually get on to automate processes or to use AI in automation, you could actually use this in better management of how you would like to go about day to day activity or you know, how you’d like to structure your work process. And essentially, once you have worked on the system, for a considerable amount of time, you can also completely automate this process, thereby saving a lot of time so that you can actually apply your mind and have more time to apply your mind to the legal issues. And some of the bigger questions of law instead of actually, you know, having to spend the majority of the time just extracting and reading or reading information from the documents and being extracting information from there.

Namit: I see. Okay. So, let me summarise it. Assume that I am the law clerk here. So what the product is helping me do is first not be dependent on me to go through each and every line and make my starting point of the reading of the learning understanding, already a prepared brief, which has fetched all the relevant facts that I might need.

Manthan: A little correction there, it doesn’t make a brief for you. It basically is able to answer questions that you post to the system, right. So similar to how you have Siri, or you have Alexa, or you have Google Home, which are able to answer generic questions about a day-to-day activity and questions from the web. Here, the difference that would be able to answer questions from the document.

Namit:  Okay, so as I’m going deeper into studying the brief, I’m using a chatbot and I am able to ask a few questions. In case there’s something that I’m trying to understand, I’m able to immediately refer to the chatbot, who’s able to tell me that no, it was on that date, that XYZ incident happened, and over time, what the ways in which the daily practices management practices of my research, what I highlight, where I highlight, or you know, what kind of information I copy-paste and take into another file, it will be able to understand my own behaviour over time.

Manthan: I will be able to understand more my behaviour over time. More importantly, and I will have the power to put something on autopilot or not.

Namit:  Such as?  Can you give me an example Manthan? 

Manthan: Such as, you know, for example, while preparing for a criminal case, one of the most, I mean, not one of the most complex, but proper tedious tasks is that, you know, compiling all of the evidence. So, you know, if you’ve done this work, you know, four or five months, then this is something that you might rather put on autopilot, and basically, review the autogenerated list of evidence, and see where they’re coming from, and just verify whether they are correct or incorrect.

Namit: How would this happen in practical terms? How would that list of evidence be generated? Or when you say you can simply autopilot it? How do I do that?

Manthan: So on the system itself, we’ve given frameworks where the user can actually extract information from within a file and tag it as evidence, or tag it as a witness statement, and give more subtags, as they would like to structure it. And the system would understand these structures and show it to the user. And when they’re showing it to the user, they’ll show to the user that you know how many times have like, say you’ve done 50 annotations, which basically 50 extractions of information from various places from one document, and this, you’ve done this task you’ve done in around 100 files. And it is expected that the AI accuracy to automate this tool, basically, we have been able to identify a pattern, which is about 80% reliable. So at a click of a button, you’ll be able to test this autopilot, put it on, you know, put on automation, and be able to see if the computer’s 80% accuracy is actually 80% accuracy, fantastic or not. If it is not, then I just give the system of feedback where I do not agree, or I agree, or rather, I just go about deleting that information and adding the information that I think is relevant. And the computer learns again.

Namit: Yeah. Okay, fantastic. So in. In this particular use case that we were the scenario that we were talking about the user will be you already mentioned the law researcher or the law clerk? And is there a specific reason why a judge can do that?

Manthan: The judge can also do that. Oh, just that from the ways that we’ve seen it is usually the law clerks who prepare the first briefs. Then brief the judges about, you know what the case is about all, all the judges do simultaneously read those matters themselves and create their own notes and highlight separately.

Namit: Okay, so when the word gets from the law clerk to the judge, then does the system have any utility specifically for the part that only a judge is doing?

Manthan:  Could you repeat that question, please? 

Namit: What you were saying was, you know, I’ll just summarise what you were saying that all of these activities, all of this analysis, reading is typically being done by the law clerk, which is why it only stands to reason that that would be the user in 80 90% of the cases that would be the user undertaking all of these activities. Does the system continue to be relevant when with the judge when it gets handed over from the clerk to the judge?

Manthan: Yes, because the activities are done, whether by a law clerk or the judge, the activities are essentially the same, which is reading through the document whether you want to do an index search, to get to the relevant information, or you know, you highlight some information, you tack some information or you drop some documents, whether you’re drafting questions that I want to ask him the next hearing, or whether I’m drafting, you know, a chronology of events that are happening. But these are all common activities, whether they’re conducted by law clerk in the case of the Supreme Court, or whether they’re conducted by a judge himself or herself in the case of the majority of High Courts.

Namit: Got it. And what’s the status? Has this been more mobilized or operational in the supreme court?

Manthan: In the Supreme Court, currently, they are procuring the hardware because of privacy concerns, they wouldn’t I mean, you know, we wouldn’t run it on our hardware. So, the Supreme Court as a data center and it is procuring necessary hardware to run such powerful AI systems because processing natural I mean, we are doing multiple layers of processing here, I mean, the first conversion of scanned images to computer-readable text. Even within that, there are about 20 or microservices of, you know, when you talk about noise removal when you talk about stamp detection, signature detection and separation, we’ll talk about document indexing. So there is heavy computation at each level, which requires heavy hardware. So Supreme Court is currently in the process of procuring the hardware, after which it will be fully operational for everyone to use

Namit: Got it.  I understand regarding the current status, where it is, and possibly I can imagine the few next steps. If I can just go on a macro, macro-level in terms of the justice system, and expediting how expediting is still a specific kind of innovation. But if you’re increasing the efficiency of the system, in general, let’s start with just disposal of cases, what are the next set of systems that innovators should probably look at that we need today, but we don’t have them what are some very, you know, like, severe pain points, where there is a need for innovation over and above what SUPACE  is already covering?

Manthan: I think there is a very strong need for prediction models, right. So what we are doing right now with SUPACE, only process automation. However, we are also coming out. I mean, we are also doing R&D for a product that is being designed for lawyers and corporates where we are also exploring prediction models. And I think this is one area which is very important and essential for all sides of the stakeholders involved in the Justice delivery system, whether they’re the litigants themselves, whether they’re the lawyers, or the judges and the courts. 

Just to give you a simple example, you know, I was looking at one of these cases, data back in 2005, or six, I don’t exactly have the date on. But so so this was, you know, a case where it was assigned to, I mean, it was handed over to an arbitrator. And the arbitrator had given an award, saying that, you know, the amount of five CR or somewhere around that is supposed to be given by one of the parties to the other, and the other party who lost the award, the end of appealing this matter in the High Court. And, again, there the decision came in the same consistent direction that you know, the board is supposed to go to the party, but by this time, the value that was to be given was somewhere upwards of 350 crores with the compound interest. So, now, this is a pain point, not just for the litigant but also for the courts, you know, where about, you know, a fair decision has been made, how likely is it that this decision would be overruled? How likely is it or rather, for that matter, what kind of case laws can be presented for and against any argument or fact or an issue that arises in a case? This is something that is important for all sides of the stakeholders involved. Although this tends to create bias because whatever is in the majority, or whatever looks good for your scenario is what you will end up using. But that is also how the industry functions today, right? I mean, you go to a research portal, you research case laws that are going to help your case, you know, help you make your case stronger. But at the same time, if you have the power of AI to aid you in this area, you’ll be able to do it more effectively.

Namit: So, you’re able to the need to be able to make informed decisions about even starting on a part of litigation, or arbitration, before you do so. And I think what you’re saying is, it’s not like this is anything, anything new, you’re already making these judgments and decisions all the time in your head, advising your clients on the basis of whether it makes sense to go in this direction or not, it’s just more accurate, if you’re able to use the power of machines, to be able to take things a little deeper and give you a prediction, which is more accurate of the real-world scenario than you will process in your mind.

Manthan: That’s true and this is truer for bigger organizations, such as you know MMCS who have heavy litigations or various government departments, where in some cases you know, the allocated budget, it falls short of what has to be given compensated as a part of the litigation is being done by their department. So these are policy decisions, these are, you know, management decisions or these are just strategy decisions that are useful for any litigant whether I do I have to do I want to litigate, right, even courts want to reduce pendency. When courts want that, you know, if there is out of court settlement, then it is in the favour of the courts because the courts will then get to deal with the cases that they are already being huge pendency of so this is something which is beneficial to all, I believe more predictive models will help make everyone make better decisions. And, you know, we’ll help unclog the system, as it is a seeming clock right now.

Namit: Yeah. If I can look at justice med administration as a market for innovation. There’s another, you know, there are now developing a parallel private court system, ODR systems. Is there a scope for the same kinds of processes to be applied in the parallel system as well?

Manthan: Yes, because the processes are the same, whether you’re doing it out of court settlement or whether you’re doing it in the court or whether you’re doing online dispute resolution or you know, you’re doing dispute resolution now with the help of an arbitrator, you know, the processes are essentially the same. You know you have to read those files, you have to accumulate those facts and issues and find relevant case laws for them and be able to ask the right questions and be able to answer them in the most as they come.

Namit:  As a company, would you be looking in that direction? 

Manthan: There are already some people working in online dispute resolution, we definitely be looking to partner with those companies in order to make those platforms even more effective than they already are. 

Namit: Yep. Okay, and with respect to the implementation of the systems in various High Courts, over time in the district courts, and Supreme Court, is it essential for us to have a unified system, so that cross-pollination of processes say that there’s an appeal that goes from one court to the other, we can use the efficiency in the previous system, have the same standard set of processes, and then, you know, like, benefit from that at an applet level?

Manthan: So, yes, it is optimum, if we do that. However, in the real world scenario that rarely ends, right. I mean, even if you look at the E-filing of cases, I mean, e-filing, AV court, has developed its own e-filing. Whereas, you know, it could be easily like one corporate has developed it, and every other court could have used it. But that doesn’t really happen in the practical scenario. There are many reasons for it, I’m not an expert in that area to actually say why that is happening, or why that happens. But yes, it is definitely very beneficial, highly beneficial, if a single unit like you know, a common system is used across the country, there’ll be more consistency, there’ll be more transparency, and more ease of operation overall, even when documents are being transferred from one court to another court, let’s say from a high court or Supreme Court, or from a district court to a high court, it makes it much easier if it’s a single system that is unified across the country. 

Awesome, that is, I mean, I believe that is a good direction, I mean, all countries are also anyways going towards the right. India has started moving in that direction with the GST movement, one nation one tax. And we various other things that are just trying to do with E passes and other areas that are trying to unify the nation into a single system so that you can both centrally and in a fragmented way, look at the problem statement and address them critically.

Namit: Okay, so would it be too far-fetched to say that ultimately, one direction where we can imagine going is, for example, how we have made a single payment system, and then developed an India stack? And now we are able to power developers to be able to build on that, and innovate and global solutions? The direction in the justice system could also potentially be that?

Manthan: Yes, we are hoping for that. We’re hoping for that. Instead of multiple payment portals coming and competing against one another, it’s better if one portal is spread across, and then multiple innovators come and innovate in separate areas. And the system overall, I think that is definitely the best approach one can take.

Namit: Fantastic. Okay, Manthan, where are you heading in the next five years or so?

Manthan: So in the next two to three years, we want to focus on the legal fraternity. We’ve already started with the judicial justice delivery system, we are already in talks or trying to initiate conversations with various government departments where they could leverage a similar or same technology in order to fast track their, their processes. And at the same time, also leverage the power of AI and already developed algorithms in order to identify as I was giving you the use case, whether they should go for litigation or not. Right. Or settle out of court. So there are these areas that we’re looking at right now, in the legal space primarily. Also, we have inquiries coming in from other countries, both the judiciary of other countries and the law firms trying to inquire if the product can be used their language if it can, you know, if you can conduct a certain process on this on this A product. So so these are the areas that we’re looking at for the next two to three years, we want to refine and make it more usable for the legal fraternity, while also trying to get hold of other areas of work in other industries and trying to develop algorithms for them. And hopefully, in the next five years, we should be able to capture one or two more areas or one or two more industries where the exact same technology or the exact same framework could be applied in order to optimize productivity.

Namit: In the public space?

Manthan: Both public and private space.

Namit: Okay, so should you be looking at the, at the depth in this industry, trying to go deeper, improve the systems by having multiple applications at the same system in different scenarios, and also go abroad? And, you know, the applications that you’ve already developed? There would be some low-hanging fruits where you can potentially apply them?

Manthan: Yes. I mean, we’re already working on this. I mean, this is already being branded as legal tech, this product, right? But you could easily just reconfigure some of the areas of this application and make it usable for a different industry altogether. For example, auditing firms or insurance companies, banks, NBFCs. These are all industries, which are high-value services, You know, which have suffered heavily due to post-war in the post COVID scenario, where working from home isn’t really just as effective as it was working from offices. And at the same time, you have a high pool of resources to do repetitive tasks. And, and yeah, so these are the industries are looking to venture into in the next five years.

Namit: Fantastic. Awesome. Manthan, how and where can people reach you?

Manthan: So, reaches from the website, LinkedIn? Yeah, good. These two are the best ways to reach me.

Namit: Personally, and, and your social media handles? I’ll provide those.

Manthan: Yeah, yeah, I’m usually I’m not very active on social media. 

Namit: Why is that?

Manthan: Long before the social dilemma came. So when I was in the US, I observed this pattern that you know, that lots of data are being collected. And every time, I browse for something, I start getting ads, you know about that, or if I’m even having a conversation with a friend on a phone. I started getting ads for that. And it was just a little unsettling for me. So I, I stepped, I took a step back and became a little dormant on my social media handles, except LinkedIn, because which is a work handle and I’m usually more active there.

Namit: Fantastic. Okay, thanks Manthan. Once again, thank you for your time. 

Manthan: Thank you.

Namit: Loved speaking with you. Thank you.