Joe Green (TechForge):

I’m joined today by Adrish Sannyasi from Rhino Federated Computing Platform. So Adrish, welcome to my little corner of TechEx North America. Thanks for spending a bit of time with us. Now, we always start these interviews with the easy questions. The first one is always, you better tell us a bit about you, where you came from. I’ve done some research, but obviously people watching this video might not have done so, so you’d better give us a potted history of who you are.

Adrish Sannyasi:

I’m originally from India. I graduated in engineering and then I came to the US a long time ago now. So I worked in various industries, started as a software engineer, working on a lot of Oracle related applications, databases, Oracle applications. And then I worked at Oracle Corporation as well as a sales engineer, solution architect, in healthcare and life sciences mainly. But then I worked at a startup and then went to Google Cloud in the early days of Google Cloud and worked with a lot of lighthouse customers of Google in a healthcare and life sciences space, and helped them with the cloud migration data and AI infrastructure and helped them implement early versions of a lot of AI products. And then I joined Rhino back in 2023. And then from there we build a team of solution engineers, forward deployed engineers, customer success managers, and deployment strategies, which is the technical programme managers. So we have a team that supports commercial organisations and helps get value from our products.

Joe Green (TechForge):

And I think you’re downplaying some of your background here. I mean, it kind of reads quite impressively, if you don’t mind me saying so? I mean, there’s a mention of a little school called Stanford. You’ve got graduate level diplomas.

Adrish Sannyasi:

So I went to medicine first of all, I went to business school to study MBA at the University of Maryland and then I studied biomedical informatics.

Joe Green (TechForge):

This is your specialisation, isn’t it? Healthcare, sciences, data science, but in the ‘getting humans well’ sphere. This is your specialisation.

Adrish Sannyasi:

Exactly. So I went to study back in 2012 when healthcare AI was very in the early stage to study EHR data, imaging, genomics and so on. So that gave me a lot of good foundations that helped me to navigate this industry in the last fifteen years.

Joe Green (TechForge):

So I’m assuming then that when a potential client or an existing client for Rhino comes along and they’re in that sector, it’s “here’s Adrish, he’s our expert.” And you speak their language, I guess already.

Adrish Sannyasi:

Yeah, absolutely. And also now, one of the things I found is once you solve a problem in one industry, like healthcare [and] life science are very complex, some of the data problems are actually applicable for other industries as well; like, manufacturing: We’re finding problems in utilities, the energy industry, and in oil and gas. So I’m seeing that. Previously I thought, healthcare has a data problem – it doesn’t look like that. Everybody has a similar problem. Data types are different, their business processes are different. But I’m finding that the similar semantic and syntactic issues are applicable pretty much to every industry.

Joe Green (TechForge):

Yeah, I can see that. I can also see the way that bioinformatics and in particular, individual patient data, that’s a real crucible, it’s a test bed, for the way that you treat data, in that, you have to be incredibly particular and incredibly careful. And so therefore picking up those healthcare workflows, I can see how you can drop them into, for instance, the financial sector, where they have similar concerns about data provenance and data governance and issues like that. So as well as some of the semantics, [you’ve] also got the legal and the governance issues…

Adrish Sannyasi:

Yes. Now data is becoming an important of any industry now because previously people ignored the data. They just work on the software and hardware pieces. But now because of AI, data is important in any industry because that’s really the failure [point] that will drive the AI scaling’s loss, in all the industries that we are in.

Joe Green (TechForge):

Okay, now for the boys and girls at home, as they say on TV, a phrase probably that might need a little explanation, which is the Federated Learning Platform. Tell us a bit about what federated learning is, because it’s been around for a while, and these days we’re very fond of talking about contextual data layers, but Federated learning platforms and federated learning data has been around for a while. Tell us quickly what it is.

Adrish Sannyasi:

Yes. So I’d describe the Federated Computing Platform as a combination of three different technologies. One is we call it edge computing, which is running things in a silo environment. And the second one is federated learning, which is creating models based on the data that is not centralised. You can create models based on decentralised data. The third technology is security and privacy computing – privacy enhancing technologies. So these three individual technologies are there for the last ten [to] fifteen years. But what we combined is all these different technologies to create a product that helps do this decentralised execution, but controlling the execution from a central place. So that’s what we call federated computing: It’s a centralised orchestration and aggregation plus decentralised execution of different types of workload. It could be running some data processing, but running some real time or near real time analytics, or could be doing model training, model inference. All of these workloads can be run in multi-cloud or on-prem or edge environments.

But [they] can be controlled from a single pane of glass or single kind of orchestration engine. So that’s what we call Federated Computing. And this is something deployable as a product in an enterprise environment, not something [just] in research anymore. We can combine all the technologies to create a product that creates value. That’s what we call Federated Computing.

Joe Green (TechForge):

And you’ll have to forgive my rather naive way of looking at this, but am I correct in thinking, let’s say for instance, I have a research project in an area of medicine? So what I can do is I can grab, obviously with their permission, I can take data from four or five different research institutions around the world and have that centralised, I can also take live or near-live data from hospitals around the world or pharmaceutical companies around the world and take all that data and work on it. Great. But of course, all these institutions have their own different methods of anonymisation. They all have their own policies and they all have their own governance structures and their own cybersecurity issues. And actually getting all those different bits of, well, ‘chunks of data’ together, putting them in one place and presenting them as a whole for everyone to work on, or one institution to work on at least. That’s the challenge and that’s what you guys offer?

Adrish Sannyasi:

So think about while you’re working in multiple institutions, multiple data silos, there’s a lot of heterogeneity as you mentioned. The security policies are different. They may be in a different cloud or different on-prem environment. They may have different data sources because some people may have a network file system, other people may have databases and maybe you have a cloud data storage, cloud storage systems, etc. So how do you work in all of these heterogeneous environments, but still present the data in an uniform way so that, as a researcher, I don’t have to worry about where the data lives, how the data is actually stored. I just use this data using a single set of APIs. And that’s the value we provide. Instead of the researcher figuring out how to work with five different environments, five different data formats, or security policies, you present a uniform API that addresses all the different types of security and privacy needs for all the institutions. As a researcher I don’t have to worry about that, I just write my own program using the familiar language that I [know], maybe Python, R or other [language]. And maybe nowadays I don’t have to write anything in Python anymore. I just write using English. And then everything magically happens behind the scenes.

Joe Green (TechForge):

So you don’t have to worry about grabbing from a NoSQL database, and [a] Postgres database and network file shares and Azure pipelines. It just all happens and you can be a researcher. Brilliant. OK, so now I’m going to go back to my imaginary health research project. OK, so I’ve got this mad notion that I’m going to pull all these different data sets together from all over the world, and I’m going to work on that collected data. Now, all the different institutions around the world have cybersecurity teams, they have legal teams, they have data protection officers, and they all speak different languages maybe, but they all speak different business languages. They all have their own different concerns.

When you work with clients, you must be very much aware of the type of business function leaders that ‘want in’ on the discussion. You must have answers, therefore, for a cybersecurity team and answers for the data governance team. Is that something that’s part of your everyday working pattern, liaising with these different functions?

Adrish Sannyasi:

Yeah, so one of the parts of our sales cycle, or the post sales effort, is to work to coordinate in different types of organisation, in the single customer or multiple customers. For example, maybe the researcher or the ML engineering team has a project or they are working on the program. But they will be working with their cloud team, there’s a security CISOs, legal, privacy; we bring everybody in the same room, or, individually work one-to-one with them, answering questions. We get the security questionnaire, we fill those in, but also go into one-on-one discussions with all of these individuals, making sure those are aligned with the goals of the researcher or [the] goals of why people are doing this; explain to everybody. So we develop this muscle inside the organisation, so that’s part of the process. But also we have technology to address some of those things as well. Like we work on defence: in-depth technology. That means, you have security, multiple security layers in the system.

We have the physical security, like encryption, encryption at rest, encryption in transit, but also various types of application level security, like role-based access control, attribute-based access control. And then privacy methods, like differential privacy, homomorphic encryption, and also different kinds of computational execution technologies like confidential compute, which is basically running your workload in a secure enclave environment so that nobody can access the data or the model while you are running it. So using this kind of technology to make sure that everybody’s comfortable. And lastly, what we do is, we have monitoring or observability [across the] whole environment, [to see] who is accessing data, doing what. We constantly monitor, provide reports, provide intrusion detections, but also take action based on that [information]. And the last thing I would say is that we also do monitoring of the users. Let’s say somebody is exfiltrating data from the models or somebody is misusing the model. So a lot of these kind of monitoring in place in order to make everybody comfortable…

Joe Green (TechForge):

A trusted place to work. The phrase I got from your literature was a ‘data harmonisation engine’, which is a lovely concept. When I spoke to Chris Laws, your colleague, a couple of weeks ago, I think the quote I grabbed from him was, “it gets us about 90% of the way there”, because there are always going to be issues with data parsing and data sanitisation, and maybe there’s a governance issue that just, kind of needs a bit of ‘ironing out.’ Do you guys kind of help with that last 10% as well?

Adrish Sannyasi:

Yes. So the first thing is this data problem that there is everywhere. So we created this data harmonisation engine that helps you map the schema and also map the various types of terms. Like, everybody uses different terms for the disease names. We have to standardise certain terms to make it useful for different customers and different organisations here. So one of the things that happened in the last couple of years is that a lot of the models can also work with this heterogeneity. It’s not like you have to harmonise 100%. Even if you harmonise, maybe your unit of measure, or certain terms, then probably [it’s] good enough for a lot of the models, which are probabilistic answers you’re giving. Let’s say language models are probabilistic. If I’m looking for an approximate answer, 90% is probably good enough. But let’s say you’re working on a very accuracy sensitive environment. Where you are working on a clinical decision support and you need to have a 99.9% [accurate] answer. In those cases, we will work with the clinicians to do post-training of the data harmonisation model. That means you have some models that we have, but then it will be post-trained on the particular customer’s data so that it becomes much more accurate for the particular use case. That’s what our forward deployed engineers will help us do.

Joe Green (TechForge):

It sounds quite hands-on and it sounds quite, it sounds almost as Adrish, or someone like you, is in the room all the time – at least to begin with. So it must be quite labour intensive?

Adrish Sannyasi:

I would say that’s why we created this team of forward-deployed engineers. Because last 10% or maybe 1% of AI is not not productised, at least not yet. So we need to have people doing that work to make sure [the customers are] comfortable. Also one of the things is, AI is a little bit different than the software where AI actually affects your workflow – very much more than regular software. So we want to be sensitive to that; that people who are using the system who are using the software are actually comfortable with that. So that’s why we have this concept of forward-deployed engineers, so that it’s not only just creating additional services, it’s also making people comfortable with the different workflows [and] environments, so it helps create the value much faster. And it is labour intensive, but there’s a lot of re-usability of all of these methods for different customers.

Joe Green (TechForge):

You’ve been doing it a long time, you’ve probably seen it already in many cases.

You mentioned AI and agents changing the way people work. Because it strikes me that, to use the phrase again, the data harmonisation engine, that’s the perfect ([and] I’ll use the word again), it’s the perfect crucible for AI, isn’t it? It’s massive data sets that are intelligently brought together, intelligently sanitised. And now at least you have a clean place on which to work and you’re not hoping that the API holds on. It gives you a coherent basis from which to work.

Adrish Sannyasi:

Yes. So I know the advantage of having this kind of layer of harmonisation is, when a human is working with [it], they maybe intuitively know that different types of data have the same meaning. But when the agent is working with [the data harmonization layer], [it] may not understand because the agent is not that intelligent enough to understand these differences. So that’s why, in the world of agents, this semantic layer is becoming much more important right now. Because if the agent is supposed to be automating things and and working almost autonomously, and you need humans in the loop all the time, then that’s not really a good value for a lot of our customers. So having the semantic layer not only helps the researcher but also helps the agent who is supposed to do the work autonomously to help you with the job, not asking you for verification all the time. So that’s why I think that if you create that layer, it’s going to not only help the researcher but also help you with pretty much any project.

Joe Green (TechForge):

Okay, I normally, with my guests, I will normally send out questions, which I’ve done for Adrish, but I’m going to throw in a little curve ball. Do you get any pushback from old school data scientists who live and breathe R and Python? They’ve spent their careers fine-tuning their skills, and along comes a platform that, at least initially, promises, ‘you can be a citizen data scientist’. Because you can now create queries, statistical, complex queries in natural language, have your queries parsed by AI, have agents go off and do the work and pull back the type of statistical analysis that the old school [professional has] spent years learning. Do you actually get pushback from people in organisations from that?

Adrish Sannyasi:

I think I haven’t got pushback yet! People actually love that, because data engineering is very labour intensive. If you can automate the labour intensive task, people actually love it because, then, they can really focus on tasks which [are] a little bit more complex and require much more thinking.

Joe Green (TechForge):

Well, because they can go home early!

Adrish Sannyasi:

Absolutely. So I think some people, I’ll say the people [who] might be pushing back on that, it’d be the people who think that writing the R code or Python code is the job. If that’s the kind of setup you are in, then you may feel that ‘my job is going away’, or, ‘I don’t feel like I’m fulfilled’. But if your job is to improve the clinical outcome or improve the patient care, if you think that that is the job, you feel good about it.

Because now we can actually think about the patient, not think about the Python code. So I think you have to think about it. I think it’s probably the managers or organisation leaders who have to redefine the job [descriptions] a little bit. The job is not to write the Python code, the job is to improve the patients’…

Joe Green (TechForge):

…is to make people well!

Adrish Sannyasi:

I think if you think about the job that way then you’ll probably feel much more fulfilled. But because now you don’t have to worry about debugging much, because debugging is done by the [platform], you can now do much more interesting things, which [are] to define the problem, work with the customer in adopting the problem that you’re solving, and you have more time to actually think and write, and talk, [rather] than just doing the coding. I think that’s the way I look at it. It’s a redefining of some of the jobs that might be required.

Joe Green (TechForge):

Adrish you’ve restored my faith in human nature, thank you! […] One final thing Adrish, obviously we’ve been talking about healthcare and the bioinformatics space at the moment. Could you just quickly touch on some other areas, some other industries that RhinoFCP are very active in at the moment, and increasingly active, just so that we can broaden out the audience as much as possible?

Adrish Sannyasi:

You know, RhinoFCP is a horizontal technology. We have vertical-related abilities like data harmonisation for healthcare and so on. But most of the product is horizontal. So we are getting used beyond healthcare and life sciences, in financial services, in fraud detection, anti-money laundering detection and also [the] sharing of insights in real-time so that all the banks can see the [issues] and then quickly act on that.

Similarly working with cybersecurity, where the security patterns could be shared quickly in the network so that people can take actions much faster. Then, we’re working with transportation, manufacturing companies in predictive maintenance, getting federated learning into device settings and improving the predictive maintenance models. We’re also working with supply chain and retail and other settings where data sharing in real-time is important, or insight sharing in real-time is paramount.

Joe Green (TechForge):

It’s fantastic. It’s also great to see something that has come out [technology that’s] a positive for mankind, as it were. And also great to see technology being used in such a positive way; not an AI that’s being used in order to encourage us to buy more things! It’s great from a geeky, technological point of view, but it’s really rather dispiriting. So a positive message to end it on. So thank you ever so much, Adrish, for joining us today.