Privacy, Data Exchange & Public Good with Novoneel Chakraborty

Hari: Hi everyone, welcome to another Solvesustain podcast. Today I’m really happy to have Novoneel Chakraborty on board with us for this discussion.

Hari: Novoneel works on data systems for public good. His work sits at the intersection of data exchange, privacy, digital public infrastructure, healthcare and responsible AI. At the Centre of Data for Public Good at IISc, Novoneel works on ways to make sensitive data more available and usable for governance, research, innovation and the delivery of public services. His research spans areas such as differential privacy, secure de-identification, trusted execution environments, and privacy-preserving data quality assessment. In practical terms, this work enables valuable data from cities, health systems, mobility networks, sensors, and public institutions to be shared and analyzed safely while protecting individuals, institutions, and communities from privacy risks. Novoneel’s current focus is on healthcare systems in India and how digital health initiatives can be used to improve the lives of ordinary citizens.

Hari: This involves applying methods and tools from his background in privacy. So great to have you for the podcast, Novoneel. To start with, could you introduce and elaborate on your work journey and interests? What problems, especially related to public good and data privacy, compel you? And why and how did you come to work on them?

Novoneel: Thank you so much for that very generous introduction, Hari. Thank you so much for having me on your podcast. Very happy to be here. You know, if you told me when I was in college that I’d end up spending a better part of my career so far trying to make government data safe enough to use, I’m not sure I would have believed you. That’s mostly because I studied aerospace engineering. That’s really my background. I spent the first two years of my career in the aerospace engineering industry as a program manager at Tata Lockheed Martin. I really didn’t feel aligned with the ethics of working at an aerospace company. I really wanted a switch. I was looking around for what else I could do.

Novoneel: I’d always had an interest in tech. And I was interested in exploring the intersection of tech and policy in India, especially because of the very rapid digitization of services in India and the booming access to internet that Indians were receiving in the last decade or so. And so I applied to this fellowship called the International Innovation Corps Fellowship, which is funded by Omidyar Network. And it’s through the University of Chicago. And as part of this fellowship, you know, the fellows are placed with host organizations that are embedded in the tech policy space. And the expectation sort of is that there is no expectation. You go in, you spend two years with that organization, and you, you figure it out if you like it you know you can stay on with the organization and if you don’t then you can move on from the domain or from the org and I was placed with what was then known as India Urban Data Exchange (IUDX).

Novoneel: And I.. this is back in 2021 and some of the first few things that I worked on at IUDX were, data quality assessment, private data quality assessment and that sort of led into my research in data privacy. And so I think. In terms of what my motivation was, like most engineers, I assumed that the hardest part of the problem that I was trying to solve in data privacy was technical. Build the pipeline, clean the data, ship the model. But that was not the case at all. What I actually found was that, India doesn’t have a data scarcity problem at all. We have data everywhere. We have health records, city sensors, mobility networks, water systems. All of these systems are collecting, aggregating, and storing data.

Novoneel: But the problem really lies around a scarcity of trust. Nobody really wants to be the one agency, the one hospital, the one department that shares something And then, you know, reads about it in the news three months later as, you know, a breach or a leak of data. So data just sits there unused, like a library where every single book is glued to the shelf, right? No one can actually make any use of this data. So that was sort of the compelling incentive to me, not how do I build a good pipeline to process data, but how do I be involved in the creation of a system that makes sharing the data the safe choice instead of the brave choice.

Novoneel: And so that took me on this whole journey from, you know, early research work on privacy preserving methods to, you know, a government funded privacy enhancing technology productionization projects, you know, where we develop technologies like secure multi-party computation and federated learning, and sort of integrated them into health data platforms and platforms in other domains. Along the way, I started doing a master’s in data science at the Indian Institute of Science alongside my job. And that was partly because I realized I needed to somehow formalize the intuition that I’d built up on the job with like a formal education in the space. So if I had to just compress my journey into one line, I started as someone who wanted to sort of move data safely.

Novoneel: And I’ve become someone who’s now thinking about the infrastructure of trust, that makes moving data the default and not the exception.

Hari: That’s a very considered and thoughtful approach. And I think really the picture which is coming to my mind is of those books stuck to the shelves in the library. And the fact that getting them unstuck is not a purely technical problem and that you had a learning experience regarding this and what it takes to unstick those books and get that data shared.

Hari: So coming to the actual benefits of data sharing, right, or data, can you explain what the term data for public good means in practice to you and why this is something, you know, important to consider and do systematically?

Novoneel: Yeah, I think because data for public good, the term is actually part of the organization’s name where I work, Centre of Data for Public Good. It sort of starts to lose its meaning at some point because you hear it so often. But I think when people first hear of our organization or when people first hear the term data for public good, they picture something charitable, right? Like data as a donation, give away your data for the greater good. And I think that framing has done a lot of damage because it makes people think that the goal is access, get the data out, get it flowing. But that’s not really the hard part.

Novoneel: The hard part is actually reusing that data safely. I’ve heard this analogy. Which is on organ donation. The scarce resource in the organ donation analogy is never the actual organs. There are organs available, but the problem is that the organs need to be matched, they need to be transported, and they need to be transplanted without harming anyone in the process. And the infrastructure, the matching systems, the cold supply chains, the consent protocols for the organ donation list, that’s what makes the actual process of donation meaningful, right? Without it, you just have these organs that are going to waste. And so I sort of liken this organ donation infrastructure to data.

Novoneel: Data is the same. Data for public good in practice means building that matching and transport system, not for organs, but instead for information. So that, you know, a research institute or a city planning department or a health ministry can actually use data that someone else has collected without that reuse becoming a privacy disaster or some sort of liability for the institution. And this is what we are actually trying to do at CDPG. We are creating the infrastructure that allows these different stakeholders in the healthcare ecosystem, for example, to sort of interact with each other and move this data safely. And we’re not asking you know how do we get more data but we’re asking how do we make the data that already is available, usable by a second or third party who wasn’t the original collector without getting anyone hurt in the process of that use I think that’s what the actual actual practice of data for public good means.

Hari: Yeah, that’s super interesting, Novoneel. So I understand that the data for public good is misunderstood many times and it’s actually a process of exchanging data so that it creates value without creating harms.

Hari: When it comes to the actual potential of data exchange to create public good outcomes, can you elaborate a little bit more on how data exchanges work and the potential public good they can create in line with goals such as the SDGs or sustainable development, for example, we could choose applications from healthcare or from all across the sustainable development spectrum, if you feel that’s appropriate.

Novoneel: Yeah, I think, you know, when I joined IUDX, the term data exchange was in the name, right? And it was also, it was a reference to the actual act of exchange of data, but also to the platform, the exchange, like a telephone exchange that we used to have. This was a data exchange. And the idea was that data sets from really diverse and potentially disparate domains could be cataloged on a single platform. And the idea was that consumers and providers of these data sets could talk to each other through this platform and consumers could request access to specific data sets that they found interesting or they wanted to work with.

Novoneel: Directly from the providers. And the provider would be able to give consent to use that data that they owned or that they were the fiduciary of. And they would be able to provide consent to use it for a specific purpose or a specific duration. And then the data consumer would then go and build an application or do something useful with that data. And so the idea was that we’re breaking down these silos where you have multiple individual entities that are collecting and sort of holding on to their data, now you’re putting all of them in one centralized exchange and you’re letting people sort of choose, you know, which of these data sets best fits their needs.

Hari: So, in terms of, you know, applications, so data exchange is the modus operandi of getting those, you know, getting data where it’s needed, right? In terms of applications, do you feel that these applications are governance as usual or are they some new applications like risk ratings for agriculture, for example, for farm loans? Or do you feel that some modeling-based applications might be enabled through data exchange? Because my understanding is that the data exchange infrastructure is infrastructure, right? And then applications run on top of that infrastructure in the same way that the roads are infrastructure and then, you know, private enterprise runs on top of that or public transport runs on top of those roads, right?

Hari: So do you have anything you’d like to share in terms of the potential applications that data exchange can support towards creating public good in line with goals such as sustainable development or other national or global imperatives?

Novoneel: Yeah, I think I really like the road analogy because I think I’ve used it somewhere before, but we’re actually, we are building the roads. We are enabling the cars to actually run smoothly from point A to point B on these roads. And these cars themselves are the applications, right? And these applications, pretty broad, we’ve seen, maybe I’ll give you a few examples in the urban space. We had a data consumer. Request access to the streetlight information in the city of Pune. And using that information, he created an app called My Safetipin. And the intent of this app was to sort of provide safe walking paths for people along the best illuminated roads in the city.

Novoneel: And so it would give you a path from point A to point B, but it would follow the best illuminated path and avoid darker areas, which may be unsafe for pedestrians. That’s one example that someone did in their own private interest, but that served a public good. Similarly, in the cities of Surat and Varanasi, we deployed something called a green corridor, which was essentially a geofence for traffic signals, which would turn the traffic signal green when an ambulance would enter that geofence. This would sort of allow.. ensure that no ambulance was ever stuck at a red light waiting for the light to change or have someone you know manually change the light it would automatically turn green and then once the ambulance passes it would go back to its normal operation, and I’ve seen these actually work in practice and and they do actually do what they’re intended to do.

Hari: So it seems that there’s a great scope for public private partnership and innovation when it comes to building the applications.

Hari: Given what you’ve seen in terms of the applicability of data exchanges, right, in this regard, what do you feel are the limiting factors which prevent data exchanges and their applications from creating their full-scale public good and development impact, if you can talk about those?

Novoneel: Yeah, I think people assume that the limiting factors are technology-related, but they’re usually not right I think, the first real blocker that we’ve seen in practice is a lack of institutional trust, or the lack of a legal safety net right agencies, are not going to share data just because the policy says they can, what they want to know is you know if something goes wrong who’s liable you know who’s accountable and this is something that you know really needs to be addressed and is sort of addressed by the the DPDP Act which is going to be operationalized later this year, But this lack of trust has really been one of the main limiting factors for, you know, adoption, I’d say, of data exchanges.

Novoneel: The second I would say that we’ve really seen has been a big blocker is what I’m going to call the connector problem, right? There are so many legacy systems that are hosting these data sets or collecting these data sets. It’s like trying to plug an Indian appliance into an American socket, right? It just doesn’t work because of the voltage difference. So the data exists and the intent to share the data may exist, but the actual technical bridge, between, you know, an old hospital’s information management system or an ERP platform and this modern data exchange built on, you know, modern standards, that doesn’t exist. So we’re seeing this in, you know, many different projects, including one that we’re working on in the retail industry, where, you know, you can have multiple different data sources, ERP systems, CRM systems, you know, all sorts of different systems, enterprise buses, and all of them are contributing different types of data, different formats of data.

Novoneel: And somehow, you know, this data needs to be aggregated and orchestrated, but the adapter to do that doesn’t exist, right? And that’s one of the problems that we try to solve, which is we try to build these adapters that can harmonize these disparate data sources. I think the third one that I’ll talk about is unclear consent and benefit sharing frameworks. So recently, we did a work shop that was hosted by ICMR and our DHDS and, in IIT Bombay. And the idea was a data governance blueprint for how AI can be harnessed for healthcare data. And this came up again and again, right? It’s not that people object to data sharing in principle. It’s just that nobody has clearly answered what happens to the person whose data this is if something goes wrong.

Novoneel: Or what does the community that generated this data actually get back? What is the point for them of sharing the data or being involved in the process? Of the creation of this data. So until that’s answered clearly, I think sharing will stay frozen, not out of any ill intent or malice, but out of this unresolved responsibility. I just want to talk about one last thing, which is incentive, right? And I think this incentive problem is something that we saw recently, when we chaired a roundtable for CXOs from multiple different private hospitals. And the idea was to ask the CXOs, the digital infrastructure exists, ABHA IDs exist. The NHA has defined these ABDM guidelines for hospitals to be conformant to that enable cross-hospital data sharing.

Novoneel: But there’s just no adoption. And we wanted to understand why that adoption really doesn’t exist. And I think the answer that surfaced from that conversation was there is a lack of incentive. There is no benefit for a hospital to actually share the data that they are generating, that they are holding. In fact, they are more incentivized to not share that data because of fears of patient poaching. So the benefit accrues somewhere else, to some other agency like the NHA or to society. There is no benefit for the actual institutions involved in holding and sharing that data. And so that, I think, needs to be that pull towards participation needs to be very deliberately.

Hari: Thank you so much for the very thoughtful answer. So the key factors which you’ve mentioned are the lack of trust. And this would be probably related to my next question. But then apart from that, the conversion problem, I guess, which I think the DXs of CDPG are solving for already, right? Through the adapters. And I think unclear impacts and benefits yeah that certainly is something which probably could be clarified going ahead and from a use case perspective I guess it’s specific to each use case and it has something to do with the last point which you mentioned which is that the incentives and the alignment for people to participate in data exchange are unclear so So would you say that the incentives and the benefits, it merits a closer look at how use cases are designed with like a good appreciation of potential, you know, even private models which can run as part of the data exchange or a closer look at market incentives or other incentives which can enable people to create shared value through the data exchange?

Novoneel: Yeah, I think this is something that I think going forward will be definitely the biggest blocker. Because right now, if I talk about the hospital or the healthcare use case, there is no mandate for hospitals to actually be compliant with the ABDM protocol, right? I mean, there have been certain roundtables where the NHS said that this is how you do it. But the hospitals simply don’t have the capacity, one, to actually do that technical, work of converting their in-house EMR and HMIS systems to be compliant with the ABDM. But two, they don’t have that incentive. So I mean, one way of creating incentive is to make it a mandate, right? Like you incentivize not complying.

Novoneel: And in that way, you incentivize compliance. But I think this whole idea of benefits and incentives really needs to be given some thought. And it’s not as simple as just laying the infrastructure rails.

Hari: That’s really interesting. So how do you feel that the conversation is changing, especially when it comes to the other point that you mentioned, which is that the fallout of a data breach or something bad happening is not clear? How do you feel about this in regard to recent privacy legislation in India? And can you explain exactly why and how privacy considerations might need to go beyond a purely legalistic or compliance approach to being a part of the governance of digital public infrastructure itself? And do you feel, you know, that this can unfreeze data exchange, at least insofar as it’s frozen because of a lack of trust.

Novoneel: Yeah, so I think the conversation is changing in recent times, especially since the DPDP rules were notified in November of 2025. And now the ministry has notified us of the enforcement dates, the 14th of November 25, 14th of November 26, 14th of May 2027, and with full provisions of the Act and rules coming into force by 14th of May 2027. And more specifically, I was talking about disincentives. I mean, penalty provisions become enforceable from November 2026, once the data protection boards power switch on, with penalties, running up to 250 crore for failure to take reasonable safeguards. So I think this is really, I mean, at least what we’re seeing is like, this is a very scary number, this 250 CR.

Novoneel: And it’s actually driving a lot of fear. I mean, I don’t know if fear is the right word, but this is actually making, the conversation really about compliance. And now that there’s a deadline, it’s about how do we become compliant in the fastest way, this leading to organizations bringing up that are specifically working on compliance. But I think the distinction that you asked about is that I think compliance is a defensive term, right? It’s about staying out of trouble. And we did the minimum that the law asked for so we can’t be penalized, right? That mindset is never going to actually get anyone to share data, it just protects you if you’re forced to share data, so I think what we need to think about instead is private privacy as a generative infrastructure, something that doesn’t just protect you if you share but actually gives you a reason to share. So it’s like a, if I use another analogy here, it’s like a a bank vault versus, UPI, a modern payments network like UPI.

Novoneel: So a vault is compliance, right? Just lock it up and make sure you don’t get robbed. But you can’t really build an economy if all your money is tied up in vaults, is locked up in vaults. What builds an economy is something like a payments network where the security is baked so deeply into the system that people stop thinking about the risk completely and they just transact and they just interact with the network. And I think that’s what a real privacy layer in DPI in digital public infrastructure should feel like and I think, proper implementation of privacy protocols can unstick data exchange because. I think a lot of the hesitation is psychological.

Novoneel: And like a hospital administrator is not going to share data because the law forbids it. They’re not sharing data because nobody has made that risk of sharing feel lower than the risk of not sharing. So once you have this credible privacy layer, one that’s real and not just a policy document, then non-participation starts to look like the risky choice because everyone else is participating in the sharing of data inside this ecosystem and you become the odd one. You’re still doing it in the old way. So I think that’s how privacy needs to be more than compliance in data exchange.

Hari: Yeah, that’s a really fascinating comparison, actually. I totally appreciate that once upon a time in India, we transacted everything in cash. And then now it’s like the default transaction mode is through UPI, right? And this has been such a shift which actually worked. And it’s baked into the infrastructure, like the secure financial transactions are somehow baked into the UPI infrastructure. And what you’re saying is that in a similar way that people exchange cash and this powers the economy maybe, data exchange will also happen if it’s baked into the infrastructure one day and maybe soon. And this will power value creation of all kinds in the economy.

Hari: Since we’ve come to the topic of how to bake in privacy-preserving data exchange into the infrastructure, right? I think you have done significant work in this regard. So could you share the technological basis to how this works in terms of anonymization, differential privacy, trusted execution environments, some of these technologies which, in fact, we’ve collaborated on as well. So could you explain how these contribute to the picture which you just painted?

Novoneel: Yeah, I think let’s envision a toolbox. And I’m going to open this toolbox and describe a number of tools that are present inside. And I think the trick or the important part isn’t picking one favorite tool, but it’s knowing which tool is right for the job, which tool the situation actually calls for. And the first tool is anonymization or broadly de-identification. So this is where you know you clean the data before it actually is exchanged and at CDPG we’ve done some work on k-anonymization which is you know a way to sort of hide individuals in in groups we product we productized a pipeline around an open source tool called ARX and we sort of enhanced the underlying algorithm which involves building what’s called a generalization lattice, which is essentially a map of every possible way that you could blur or generalize a data set.

Novoneel: And you then search that map intelligently to find the least amount of blurring that you need to still protect people. It’s like adjusting the focus on a camera lens. You want, if you’re trying to preserve the privacy of people in a group, you want to put just enough blur so that no individual face is recognizable, but not so much that the whole picture becomes sort of useless. And we published a paper called Skald, S-K-A-L-D, sort of named after the Norse, a Norse form of poetry, called Scalable K-Anonymisation for Large Datasets, S-K-A-L-D. And yeah, that was our work on de-identification. But another layer that we added to that was differential privacy.

Novoneel: And this is for situations where even anonymized data is still risky. So you would instead also add calibrated statistical noise to the results of a statistical query. So it’s like a survey where instead of asking, did you personally do X, you flip a coin first. The aggregate answer will still be meaningful, but no individual response can be attributable to an individual. That’s the idea behind differential privacy. And can you… Combining these two together, we sort of produced a paper called SPIDER, which was a secure pipeline for information de-identification with encryption. And the idea behind the paper spider was that you would have this end-to-end anonymization pipeline which included not just k-anonymization and differential privacy, but also some more fundamental tools like pseudonymization and suppression and masking and hashing and so on.

Novoneel: And so you would be able to you know take this application and use it to anonymize or de-identify your data. And the end-to-end encryption part of that story was done using what are called Trusted Execution Environments, TEEs, or Secure Enclaves. And this is like, you can think of it as a sort of black box, right? The idea behind the Secure Enclave is that you are given certain hardware-level guarantees of encryption by the actual manufacturer of the chip on which your compute is being done. And these portions of the chip are sort of air-gapped from the network, and they have encrypted memory. And it’s essentially like a black box. You’ll have your data be encrypted.

Novoneel: It’ll go in to the secure environment, into the black box. The computation will happen inside this hardware guaranteed sealed environment. And in this case, the computation was the execution of the anonymization pipeline. And then the only thing that would come out of the black box, unencrypted, was the output, the actual anonymized data, never the raw data. And so using this model, you could have any entity sort of anonymized data without ever having to actually see the data that they were anonymizing and only receiving the outputs of that data. Maybe I’ll just briefly touch on a couple more technologies that we’ve sort of worked on and developed. And this is more on, instead of moving data to one place at all, you move the compute to where the data already lives.

Novoneel: And these two technologies are called federated learning and secure multi-party computation. So in this way, you can sort of have multiple parties jointly compute something like a shared model without ever having to see each other’s raw data. And this was a very central part of what we were building at CDPG. So these are the two models that we were trying to look at. One is move the data to the compute, and one is move the compute to the actual data. But I think what I’m trying to touch on is that none of these is ever the only answer. None of these different tools that we talked about, k-anonymization, differential privacy, trusted execution environments, federated learning, secure multi-party computation, none of them are ever the answer alone.

Novoneel: The real skill, and this is sort of what we are trying to build capacity in, is knowing which combination of these to deploy, depending on how sensitive the data is and who needs to be able to access it.

Hari: Yeah, that’s a really fascinating, Arsana Vanil. And as an engineer, since I’ve also worked on some of these with you and with others, I would really recommend any, let’s say, impact-oriented engineer to consider these technologies. Because not only are they technologically fascinated and fun to work on, but they are also extremely potentially impactful once they will be integrated into the digital public infrastructure.

Hari: So coming to that, keeping the discussion in mind, what could be the full-scale impact of having a well-functioning privacy layer in the DPI? Right. And does it take on more meaning, especially now when I guess we are in the middle of an AI transition and in a society potentially where AI agents will consume increasing amounts of sensitive data? Like how does a privacy layer in the DPI play a role there in not only safeguarding citizen privacy, for example, but also in enabling AI to create societal benefit and economic value?

Novoneel: I think if this works at scale, if a working privacy layer is properly implemented, then the impact isn’t just faster data sharing. The change becomes in what is now possible to even ask. Right now, a researcher who wants longitudinal health data from multiple hospitals, they have to renegotiate that trust with each hospital from scratch for every single project that he’s doing. It’s like every time you want to cross a river, you have to build a new bridge just because you’re crossing it at a different time. But a working DPI layer that is implementing privacy means that the bridge always stays up. You build it once, and then every subsequent project that you’re doing can cross it immediately. So I think that’s the real impact.

Novoneel: And you asked about AI. I think AI completely changes the stakes, right? If you have a human analyst who’s querying a database, they ask a question, they get an answer, and then they stop. But an AI agent doesn’t stop. It reasons over the data repeatedly, autonomously, across multiple sessions. It can chain inferences together in ways that the original data owner never even anticipated. And that’s a completely different threat model. It’s it’s like someone reading one page of your diary when you hand it to them versus handing someone, your diary and them having a research assistant who will read it cross-reference it against everything else that they know and keep coming back with you know new questions and new queries indefinitely, so the the privacy layer will stop being just a gatekeeper for access I think it It has to become, the thing that sort of.

Novoneel: Governs the inference itself. So what an agent is even allowed to conclude, not just what it’s allowed to see. I think we are sort of exploring this concept with some health intelligence agents, some use of agents in some of our different projects. But the idea is like, don’t just plug an agent into a data exchange, but build the privacy governance layer as the substrate that the agent is operating on so that every inference the agent makes is already constrained by what’s appropriate to know. And I think in a world where AI is becoming more and more autonomous with every new model launch, that layer isn’t just nice to have.

Hari: So that’s a really fascinating response, Novanneel. And I think you’ve really stressed the precautionary application of privacy in the DPI in a situation where AI is learning more and more and stacking up that learning. When it comes to the beneficial impact, if we think about systems like UPI or even the Aadhaar system, for example, these were pioneering, I feel, and have also been, to some extent, generalized and even exported to other places through various initiatives.

Hari: So, in your opinion, if things go well and we have a well-functioning privacy layer in Indian DPI, what is the potential full-scale impact not only nationally but perhaps even globally and what do you feel could happen if things go well?

Novoneel: Yeah, I think UPI is a really good analogy because when it first came out, it completely radicalized how money moved between bank accounts, right? Before UPI, it was a multi-day, multi-form process, and UPI didn’t actually invent movement of that money. Banks could already do that bilaterally, but it standardized the rails so that any merchant or any app or any bank could plug into the same shared system and, transactions that would take multiple days started happening in seconds and for free. And now there are billions of transactions a month across an entire economy that used to previously run on cash. So if we imagine the same shift for data instead of money, I think moving health data between a hospital and a research institution right now is like banking pre-UPI.

Novoneel: A bespoke agreement and negotiation needs to be made every time because there’s no shared trusted rail for it or that rail exists but it hasn’t been adopted. So a privacy layer in DPI is part of that rail and, it’s not just another app but it’s the standardized infrastructure that lets any hospital, any government department plug in and sort of move sensitive data the same way that any bank can. Move money today. They can do it quickly, they can do it safely, they can do it, by default rather than through special exception that’s been negotiated and and just like upi you know the the magic isn’t the technology but, it’s the invisible infrastructure that I’m talking about that nobody even has to think about anymore like nobody today is asking is upi safe to use before you know before you pay, the guy who got your hair right it’s just it’s become the bar and so success for, a privacy dpi layer isn’t, when people are talking about it I think it’s when people actually stop noticing that it exists because sharing sensitive data is as unremarkable as sending someone 50 rupees.

Hari: Wow that’s really a nice way to think about it like it becomes successful when people stop talking about it and it’s just transparent right and ubiquitous so towards You know, seeing the Privacy DPI reach its full potential impact, what do you feel are the current limiting factors, you know, specifically related to this?

Novoneel: I think there are probably a few things that are holding it back. I think standardization lag is one of them. I think we were talking about standardization lag in the health system. But this sort of problem exists in urban systems. It exists in agriculture. I mean, essentially, every domain has invented their own vocabulary for governance. And so cross-domain governance becomes really difficult. If we use that socket analogy, the plug point analogy that I used earlier, it’s like every country has a different electrical socket. And the underlying idea, which is electricity, is the same, but nothing plugs into anything else. We spend so much energy and money and resources on the translation part of the problem. So I think that is definitely one issue.

Novoneel: I think two is probably capacity. I think most government departments just don’t have the people or the capacity, to verify whether a claim that something is privacy-preserving is real or just a nice presentation. And I think that gap is bidirectional. Sometimes it produces excess caution where, you know, nobody ships anything because nobody is trusting their own ability to evaluate the risk. But it can also produce the opposite effect where very weak claims might get waved through because nobody in the room can tell the difference between something like real differential privacy and, you know, just a marketing buzzword. And I think funding is probably the third aspect, the third limiting factor.

Novoneel: All of this work gets funded in very pilot-sized, MOU-length bursts of funding. Grants love some shiny new proof of concept, but the funding of the unglamorous, ongoing maintenance and development of a trust layer once it’s actually been built and once it’s past that POC stage isn’t being funded. Right it’s not it’s not something that you just build once and forget but it needs constant tending otherwise it will fail the moment you stop watching it because adoption won’t be good because features aren’t being updated and so on.

Hari: Coming to the next question if you had three asks for anyone who might be listening to this whether policymakers or private industry or standards bodies or anybody, what would they be? And this would be in terms of supporting this work and approach for privacy in the DPI.

Novoneel: I think maybe I would position my three asks to three different stakeholders. My first ask to relate it back to what I said was one of the limiting factors. My first ask would be to the funders in the space, the institutions in the space. There is, I mean, at least we’ve seen this before, but no one’s funding the boring middle part of it. Everyone wants to fund the shiny pilot or the demo, but it would be good if the unglamorous ongoing work of maintaining something like a privacy governance layer, once it exists, is funded. So that the pilot is just the ribbon cutting. The maintenance is the actual public good that’s being funded here, not the demo, not the PRC, not the pilot.

Novoneel: I think second would be to maybe some data scientists or some engineers who are listening to this podcast. I think one of the things that we are lacking is people that are sitting comfortably at the intersection of engineering, of law, of policy. You know, we traditionally tend to specialize in certain domains, but this is a really cross domain problem that we’re trying to solve. And I think it’d be important for engineers and data scientists to sort of understand what’s happening on the policy side, what’s happening on the law side. And it would also be good to sort of treat something like privacy preserving computation as a core skill, even if you’re not a specialist in PPCs.

Novoneel: Right now, there’s demand for people who genuinely understand DP or k-anonymization or secure computation well enough to implement them. And the demand is right now ahead of the supply. So if you’re sort of early in your career and looking for where impact can be made, I think this is a great space to sort of do research and operationalize some of that research. And I think my third ask would be to just a normal citizen who’s listening. I’d say, let’s stop treating data sharing as sort of a binary process, right? It’s not share everything or share nothing. But it’s good to ask questions about how your data is protected and not just about whether or not it’s being used.

Novoneel: Yes, obviously, we should know whether it’s being used, how it’s being used, but also ask, how is it being protected in that process of being used? And I think that kind of informed friction from the public will push the system builders and the institutions to sort of properly design these systems in the first place. And this sort of pressure from the public will make the whole ecosystem, I think, improve over time.

Hari: In terms of any advice you might have or suggestions for people whether just starting their career or more experienced who would like to take up work in technology for public good or digital public infrastructure which creates public good of the kind which you have been doing for the past few years do you have any words for them

Novoneel: I think what I’d say is don’t wait until you feel like you’re at a point where you’re senior enough to work on public infrastructure. I think many people assume when they’re early on in their careers that they need five years at a big tech company before they’re qualified to touch something like government data systems. I think that whole notion is sort of backwards because you need people in this field early because you need people to be building their muscle in this field, particularly thinking about things like consent, about liability, about edge cases where someone’s life is, possibly affected by a data breach. That sort of thinking won’t exist in many of these private sector jobs.

Novoneel: So I would say stop thinking about whether or not you’re underqualified for this work. I think it’s better to actually jump into this space really early before bad habits, you know, set in. And I think it’s really important to, go deep on one hard technical thing, as opposed to, you know, staying at a very shallow level on 10 easy technical things, right? I think this sort of space rewards people who genuinely understand how something works, not people who’ve been built like 10 dashboards. This was an example, But the idea is pick one of these techniques and actually learn it to the point where you could explain it to someone who’s skeptical.

Novoneel: And because that depth is sort of rare, that is what will get you into the room and into the space. I would say, this isn’t advice, but I think even if the pay in the space isn’t the best you could get, I would say what’s more important is to work somewhere where the stakes are high, where the stakes are real. So a DPI-focused non-profit or a government-adjacent project or a research lab. In these places, the mistakes that you make actually matter to a real person, and they’re not just metrics. So that’s where you learn how serious this work actually is. You learn faster, you learn more responsibly when you know that it’s real.

Hari: Thank you, Novoneel. That was really, really helpful. I’m sure that people who are listening to this are going to benefit from that. So thank you so much. We’ve come to the end of our podcast interview. This has been super interesting, super informative, and just really inspiring as well. So thank you for being on the Solvesustain podcast, Novoneel.

Novoneel: Yeah, of course, Hari. I’m really happy to have participated. And I hope some of what we spoke about is useful to the people who are listening.

Hari: Thank you. Yeah.

Novoneel: Bye-bye. Yeah. Thanks, Hari. Bye-bye.