Think AI Podcast - Compliance, Music & AI | Ep. 6 with Nilesh Maheshwari (Yuktra AI)

Episode Date: April 21, 2026

What do music and regulated manufacturing AI have in common? More than you'd think. Nilesh Maheshwari, founder of Emorphis, Emorphis Health, and Yuktra AI, joins Dave Goyal to unpack why the hardest places to deploy AI (pharma, healthcare, regulated plants) demand faithfulness over cleverness — and why structure and freedom aren't opposites.In this episode:00:00 Why a "knowledge thread" beats a database on the shop floor04:17 The real pain signal that built Yuktra (hint: it wasn't "we want AI")07:08 Guardrails for regulated AI — why clever LLMs are dangerous10:34 The plant-worker pushback that killed fancy AI answers13:23 Music, discipline, and how structure creates freedom18:54 What most US founders get wrong building with India teams21:07 Why there's no such thing as a "best product"29:22 Curious, enthusiast, or skeptic? Nilesh's take on AIIf you lead a plant, hospital, lab, or any regulated environment and you're wondering where AI actually fits — this one's for you.👉 Subscribe for more conversations on Data, AI, and the humans behind the systems.💬 Drop a comment: Are you AI curious, enthusiast, or skeptic?---🔗 Links & Resources- Yuktra AI: https://yuktra.ai- Emorphis Health: https://emorphis.health- Nilesh on LinkedIn: https://www.linkedin.com/in/nileshmaheshwari/- Dave on LinkedIn: https://www.linkedin.com/in/davegoyal/#ThinkAIPodcast #RegulatedAI #ManufacturingAI #PharmaAI #FounderConversations

Music teaches you that structure and freedom are not opposite. The best improvisation happens inside discipline. In music, timing matters as much as notes. And in manufacturing, context matters as much as information. Today I'm joined by a guest who has spent 30 years building technology across four decades of paradigm ships. From microprocessor-based process controlled in the 90s to mobile at impetus to emperous to emorphous technologies, serving thousand plus global clients, end to Emorphis Health and most recently Yukhtra AI. An operating system built for the hardest places to deploy AI for regulated manufacturing

pharma and healthcare. Nilesh Maeshwai is the founder and CEO of Emorphis, founder of Emorphis Health and also the founder of Yukra AI. He is in Thai Indoor Board, State Council member. He is a standard C-8 participant and angel investor and a past WBAF senator from India. And in full disclosure, he is one of my closest friends. We have known each other since college. We make music together and we've been sounding votes for each other,

businesses for longer than most of us can either want to admit. Today we are talking about Yubtra, which Neelich is saying that it is born at the intersectional. of power of logic, the discipline of reasoning, and the guiding thread. Welcome to the Think AI podcast. Each week we talk about the most exciting AI research, tools, case studies and more. I'm your host, Dave Goehe, and I've been working behind the scene in data and AI for over 30 years, whether you are an AI expert, skeptic, or something in between, this podcast is for you. Welcome to the show, Nish. Thank you. Thank you. Thank you, Dave, for having me at your podcast.

Really enjoying and I have seen your podcast and I'm looking forward to great conversation. Great. So let's just get started. So you Uptra stands for your unified knowledge threat for real time assistance before we get to the acronym. Walk me through how you landed on the idea of a thread. Why it is a knowledge threat versus a database. Okay, very, very interesting questions. So a database stores information and threat connects the context.

And in a plant, people do not need isolated documents. They need the chain like SOP, machine contacts, safety rules, deviation risk, training need, next action. So real work on the shore floor is sequential and contextual. One answer leads to one other. That is why Thread felt more truthful than repository. And a database tells you what exists. A thread helps you act correctly in the moment. That's great.

And I also see Thread as a guiding metaphor, isn't it? It's not just a marketing copy word. What do you have to say on that? No, no, it is. Yes, it is. Yukhtra's job is to become the stitched layer in real time. Awesome. Yeah, I think we see the.

similar pattern in fabric engagement. Most of our clients have like eight systems, 10 systems, not a single source of truth. That's why the conversation about database versus a thread which sound pretty interesting because we are trying to create unified database as a single version of truth versus YUQtra stands for a data thread, really, not a database, where you can do some realization from Yukhtra from day one. So it's not just the data, but it's also the context of the data. Does this make sense? I think it is. Yes, it is. Okay. So I have the next question which is pretty interesting. You could have kept amorphous horizontal and comfortable, right? We are talking about comfort zone. We will talk about it in a second, what comfort zone means to each of us

and doesn't mean anything to each of us. Instead, you bet capital and time on amorphous felt. We both are 54 years old and we are reinventing ourselves on each stage. And then again on YUKhtra, two of the hardest vertical intech. So, Emorphis Health on one side, Uctra on another side, service industry, a product industry. What was the specific customer signal that you were looking at, which made you bet on this solution or a product? Okay. So we do not choose these spaces because they are easy. We chose them because the pain was real, recurring and high consequences.

For amorphous health, the signal was clear. Healthcare organizations were struggling not just with software development, but with workflow complexity, interoperability, compliance, patient impacting decisions, they need domain-aware engineering, not generic software vendors. So that was one of the key reasons why we wanted to get into healthcare, amorphous health point from that particular pain point we understood the real pain and for yuktra the trigger was different but equally sharp in manufacturing specially regulated plants critical knowledge was available on paper in pdf in people's head tribal knowledge trapped in departmental

silos the strongest customer signal was not we want AI it was Our team lose time every day searching, asking, confirming, escalating, basic but critical questions. Another signal was audit and compliance pressure. Plants are under rising pressure to be consistent, traceable, training ready, but knowledge access on the floor is still weak. So, plants are still struggling with their audit readiness, pressure on production, delivery, with quality, etc. So that's where Yukra came into picture

and that's where it was born. We do not follow like the hype cycle. We followed high friction workflows where better decision and actual matters and where we can create a better impact. Now that is awesome. Especially customer signals are always the loudest right which market really understand and reacted

and you picked up on those customer signals rather than the hype of AI you are you using AI as a platform a tool to deliver where the customer is saying that this is the problem area that you need to solve and you took an opportunity to solve it. That's great. Okay, so let's talk about logic and reasoning in regulated AI. Typically regulated manufacturing is one of the hardest places to deploy AI. Now, auditors don't negotiate. You know, 21 CFR, GMP, FDA,

WHO, EMA, HIPAA, HISA, there are so many compliance out there. And this is the specific place where LLM wants to improvise or you had to stop it. What did you put in the product to make sure it stays on the right side of the line? Meaning it stays compliant. Very well articulated question, Dave, because this is one of the biggest risk in this environment. where model sounds confident beyond the validated source material. Let me give you a practical example when we are talking about SOP's interpretation.

And LLM may try to paraphrase, compress, or improve a procedural instruction to sound more natural. And in a regulated environment, that is dangerous. So for example, if a cleaning or line cleanings procedure has an exact order, timing, and verification requirement. The model cannot creatively restate it in a way that change changes meaning. So that was one of the very big risk when we are choosing an LLM or making sure that it is

giving a right context and the right answer. So we have built guardrails around source, bounding answers. Answers must be grounded in approved content only. It should have a strict source traceability. It should distinguish between approved procedure and a general guidelines. So when we are delivering any information, it is tagged that it is from approved procedure or it is from a general guideline. So that the user get an information that this is something which is coming from a strict SOP and not a general guideline. So they understand how to use it.

In sensitive cases, it should refuse to infer. And there is one important thing is that it is role-based context-aware response design. So not everyone gets the answer from the same document because it is role-based. So people get answers from the document. They are authorized to get answers wrong. So you would not get a free-form answer. Or if you get a free-form answer, it is tagged that it is from a general guideline. or a general knowledge.

So in regulated plans, the model does not get rewarded for being clever. It gets rewarded for being faithful. Now, that is awesome. And this is good for the listeners to hear. Because most people think or would think that an LLM model is pretty generalized. It's a cookie cutter.

It reads off of the internet and provides you the answer. I think it's not the model. It's how it's been designed, how it's been implemented. and what guardrails and governance I would also add is being taken care. And once you apply all those things to the right model or multi-model even, you're going to get an amazing AI or the solution which will consume AI. Correct? Yes, yes, correct.

Okay. Okay, so that leads me to another question here, which is, walk me through one specific design decision in YUFTA, that you made because a plant worker pushed back, not a roadmap item, not a customer ask, a real friction moment which changed the product. So we're not talking about the product backlog here, but really the real life pain points and the problems,

which you thought, oh yeah, this is the thing that I need to have it in YUKTRA. This is how it's going to be solved, maybe with AI or without AI. Okay, so in a real plant, and when we are dealing with plant workers. One real friction point was that workers do not want fancy AI answer, but on the floor long-form responses create friction. They want exact next step fast and in a usable format. So the real pushback was, do not give me paragraph.

Tell me exactly what I need to do. Because earlier when we started, we started building something, it was very fancy, AI giving answers, etc. But when we implemented in a real plant and the first feedback we received was that we don't need these big answers. We need one, two, three, this is the step we have to do. That's it.

Don't give so many things which is not usable because they don't have time to reach so much of content. Another thing is that it has to be real-time usable, multilingual support, audio support, because they want, while working, they want to ask a question and they should get an answer where the AI or the system Yukhra is speaking to them in their language. So that is one very important thing. And as we implement it in various geographies, even in India, there are so many different languages in which workers' workers' offices.

So we have to support multilingual. It should be easy to understand for the workers in their language. So even if it is writing a text or even if it is speaking in their language. So that was one of the important things like the usefulness beats intelligence. Yeah, this is an awesome angle you're saying, and this is how I'm translating, which is not AI first, but workers. first but worker first. So worker first using a tool to get the answers that he needs to solve his day-to-day problems, right? This is amazing. So let's talk about some of the personal things that

shaped you and I both. We have one commonality over so many others. We both came from music. You sing, I play, I sing a little bit. We've even made things together. Now music has logic, patterns, improvisation and the thread that ties it all together. And I'm very specific about the thread because it's connecting the dots and what you're trying to do at Yukhtra as well. Does that lens show up in how you architect Yuktra or even run your teams at Emorphis and Emorphis Health? Or is that just something that we would like to believe?

Very interesting and relevant question because music, you know, that music at our heart and everything would do because that is part of our day-to-day life and the way we operate and the way we work because as an artist so being a businessman is a different thing but being an artist changes a whole lot of your perspective so yeah music teaches you that structure and freedom are not opposite the best improvisation happens inside discipline in music timing matters as much as notes and in manufacturing context matters as much as information. So that is similar in yutra.

We use AI, but inside a structured environment where context, rules, source fidelity matters. So there is a definite kind of melody, rhythm, we can see in yuktra, the way we deliver things. So like music teachers layering, melody, rhythm, harmony, texture. And in yuktra, we think similar. compliance, operation, training, safety, equipment knowledge must work together, not as an isolated piece. So it is all in cohesion and everything of that is creating a very beautiful music for the listener or for the user, for the plant workers.

And it should be a very rhythmic so that they can work day to day in their plant life, in their day-to-day working rhythm. So there is alignment, listening, responsiveness, people knowing when to lead and when to support. So this is how music is aligned even in the leadership as well. So we don't want noise. We want alignment. We want responsiveness. We want people to lead as well as to support. So it also reinforces one important idea.

Like, reputation is not boring when it creates mastery. So that is true in music and in high quality operations as well. So yes, I do think music is part of the way I build good systems, like good compositions, create clarity without killing expression. So music is everywhere and everything I do and you do. I know that you are a great musician and we have been like in the same journey, sitting together on the class bench, writing music, singing and all that. And you can definitely relate to this thought process very much.

You've beautifully said it. And that just prompted me to something. So, you know, Indian classical music especially, it is not linear. It has several rules, context, you know, timings, location, modality, mood, etc., etc. So there's a structure which is also loose. I compare this with AI, which what I'm thinking right now, that AI also has similar things. You don't have predefined rules. You don't have, I mean, you could, but that will be pretty limited.

So you put rules where needed, but then you give the liberty to it so that it can come to the real life. Example, when we go and perform on a stage and things change is on the fly. You know, your mic is not working or there is a background noise or something, and then you adapt to it. So no matter how much structure you put in place, and then you also react to the people who are listening. So the song you are singing or playing may not react to the audience. And same thing for AI has to do, you know, a worker versus a manager versus an executive. When they start communicating with AI, AI needs to take a different soul and identity and start reacting to what and how they are. behaving. But you beautifully sit and I love that connection on the music to AI man. This is

amazing. And this also prompts me for another question. Thanks for asking that question. Thanks for asking that question. Because that actually like triggered some some things in my mind and as well your mind as well. Because how we can connect music with everything. And without realization, we have been living that. But when you ask, the question it actually came up in a verbal time. Yeah, there's a lot of common grounds and somehow we are subconsciously using that skill to build what we are building today and yeah, this conversation is extracting that thought out, which is pretty amazing. The other differentiation I see between you and me and a lot of

founders here in US, generally speaking, what they get wrong about building, India teams. Now I'm switching gears more towards amorphous health and amorphous so you build with India engineering teams shipping to US regulated customers and you've been very successful I see it we work together on so many of these most US founders who try to do this get it wrong and they get stuck in this compliance and or they get stuck in the cost what do they miss that you did not See, it is like I would say we have done a co-development here where we have people onboarded from the regulated industry. These are the people who have been there done that and they have been part of production, they have been part of quality assurance, they have been part of compliances, audits, etc.

So these are the people who are guiding us in actually building and shaping this particular platform. So and we are not like waiting to complete the product and then releasing it. We have been like it is a continuous development. We go and test with the customer, get early feedback and then keep on reinventing, keep on sharpening the product features as well as removing whatever is unnecessary. So we have scrapped a lot of features which we thought as engineers or as product owners are necessary. But when we are in the plant, we work with workers, we feel that some of the features are fencing item for them. It does not add more value, much value to them.

So working with them hand in hand made us realize a lot of features which are not making sense. we remove them. And I would say it is a co-development exercise which helped us in delivering the product, Yukhtra, and working with the plant owners, the plant workers, the QA managers, etc., has helped us in shaping this product. Now, that's awesome. And, you know, this is one thing to pick up on

is keeping it really lean. and outcome driven rather than keeping you with a lot of fluff and going into a cycle. So where needed you put compliance in the right order and shape, where needed you'll speed up the development, reuse the skills where needed and build a solution or a product, which is cost efficient as well as really compliant and highly secured, providing the right outcome in the right value. I want to add one more thing here that where people miss is that they want to create a best product having all the features in one go, and they keep on developing,

developing that product for long period of time. While in that process, a lot of relevance of that product gets lost. So you have to just get into the market faster, test it, and keep, this is a cycle. You have to keep on doing it, rather than creating one big, fat, best product. And there is nothing like a best product.

It is a journey. Yeah. Like you said, a basic mantra is I ship, then improve, rather than keep improving and then ship the final, which will never be final in any case. So this is amazing. So I have a few surprise questions for you also. One is first move for the leaders. You've been leading in so many ways right from your job profile into Emorphis to Emorphis Health and now into Uptra. and then going into different things like Thai board members and some of those other amazing

clothes that you have. What would you say to the leaders who has two angles? One is regulated. So regulated leaders, as you know, are always skeptic on technology, especially AI. They want to say, oh yeah, I want to do something with AI. But then they have two problems. One is, I do not know where to start. Or second, I cannot trust on AI.

What do you have to say to those leaders? So one important thing is that AI is here to remain. So it is not a fad. It is something very real. So and it has real advantage if people use it with caution with pinch of salt. Because it is not going to solve each and every problem of yours. But definitely it is going to solve a lot of. of your productivity problems, your automation, but you have to start doing it, experiencing

it, experimenting with AI. So take small step, be open. It is not going to like, as it says, it is going to heck all the system, it is going to do havoc, etc, etc. You have to be very cautious as it happens with every technology in past. We have seen. This is going to change everything. and jobs will be lost or whatever.

But this is definitely going to improve. And for enterprises, I am seeing so many use cases for manufacturing, so many use cases where people can improve their productivity, the response time to their customer, the quality of their product. AI is going to change the way you are doing business. So definitely you have to embrace this sooner or later.

It is better to be sooner in adopting the, the technology and experimenting it. Start small. Okay, no, that makes sense. And a lot of times people think about a pilot and I guess that could be a good way, either pilot or a POC, try it out rather than making an assumption on a technology and see what works and what doesn't work. And then you can make a decision because every technology comes with its own baggage and comes with its own features. And then you need to find a right tradeoff and a balance. by trying it up. Great.

I want to go back to our connections on music and college again. So we've been friends since college days. We still make music. We still call each other when things are hard. And not just on music, by the way, right? So we are sounding board outside your company and outside my partners and companies. And, you know, you could choose advisors. I could choose advisors.

Or you could choose board members. How this is more personal. but I wanted to put you on spot on that. How this experience helped you, I can talk on my part later, but how this experience help you out in what you're doing today and how you're progressing today. You know, a friend who's also having a commonality in music and being a sounding board, how that combination works out for you.

Absolutely. So as it is always said, that it is very lonely at the top. and there are very few friends where you can open up, where you can talk your heart out, and where you can share everything, like business, problems with the business, everything. And when there is a commonality of music, where it is, it is something which is connected heart to heart,

very emotional connection between us. We go around 33, 35 years. back in past when we were like in college, 18, 20 years of age. And so, and the friendship of that age is really very, very close to heart, important for you. And when you have, and when you go that long and you are in common business, you have common problems to address and to discuss, it is always a very, very good sounding board, very good friend, advisor to be on your side, to listen to your perspective, correct your course. So it is always very much valuable. You know, that's amazing. And I could say that from my side as well, that first of all,

having a long term friendship and also being in a common threat like music, we have a similar thinking philosophy, but we are still different people. deal things with differently. So when we are sounding good to each other, there's a different perspective comes through, right? From you and from me. That's one thing. But the second thing is a friend will not hold back

because he will always think about the other friend's benefits more than thinking, how would I look, you know? And that really helps to listen sometimes getting feedback is really hard and from anyone. but once you hear it from a friend, you know it's coming from a good place and when it's coming from a good place, you know where to make corrective actions. Not always you have to make it, but it does give you a very good perspective on what to correct, why to correct, and how it can take you further.

And a friend will always be more happy if the other person is growing faster than what it should be. So I want to take this platform and acknowledge that I've been, getting some amazing feedback from you and I cannot forget and thank you enough for entire life and hopefully we can be the sounding vote to each other and continue to make a lot more music together. Likewise, Dave, it is really, really amazing to be in that close connection with you, with an emotional bond, we have been with each other in all ups and downs of our life, business, college days, building music together, singing on stage together. So it has a different feeling and bond and thank you for being there.

And I always reach out to you whenever there is something I want to discuss and you're always available, present and present with a lot of new ideas, perspective. So we have corrected our course many times together. So thank you, thank you for being there. It's mutual and thank you again. I just have one question for our audience that generally resonates well with them. So the whole point of building a think I podcast is to cover three types of people. one who is AI curious, the other one who is AI enthusiast,

and the third one which is AI skeptic. Which one are you? While I know, it's good for our audience to learn who you are and why you are like that, whether curious, enthusiast, or skeptic, or somewhere in between. So I am AI enthusiast. I'm not skeptic about it. Being an engineer, being in the technology, I understand the pros and of AI. So I do not approach it with skepticism. I approach is from the angle of possibilities

and opportunities. And definitely I'm very excited to see how AI is going to shape everything, whatever we do in our day-to-day life, in our business, the way we operate, everything. So I'm very enthusiastic about AI and definitely I understand the nitty gritty-rities behind-the-scenes things of AI so I have my own perspective about AI but huge possibility huge upside in everything we do but definitely everything comes with with its downside So we have to be very thoughtful about how we implement AI, thoughtful how much we open up to AI,

like opening up all your systems, your emails and everything. So you have to still have guard rails around what it can access and how it can access. But I'm very enthusiastic about AI. That is great. Nilesh, thank you. Thank you so much for being on the show. That's Nilesh Maheshwary, founder at Yukhtra AI, emorphis technologies and emorphous health.

Find Yuta at y-U-K-T-R-A.I. I will put it on the caption. And Emorphis Health at Emorphis Health. If you need a plant, a hospital, a lab, or any regulated environment, and if you are wondering where AI fits, Neelish is your friend. start with the knowledge already locked into the walls that he has under Yuktra

and this is where the thread starts. Thank you again. Thank you, Dave. Thank you. You have been listening to Think Yeah, podcast with Dave. Take one idea from this episode and turn it into action.