skip_previous play_arrow skip_next
00:00 00:00
playlist_play chevron_left
  • Home
  • keyboard_arrow_right Featured
  • keyboard_arrow_right Validation
  • keyboard_arrow_rightPodcasts
  • keyboard_arrow_right Enhancing Compliance and Efficiency in Pharma Manufacturing with AI [Vivek Gera]


Enhancing Compliance and Efficiency in Pharma Manufacturing with AI [Vivek Gera]

Vivek Gera June 19, 2024

share close

Yan Kugel⁠ is joined by Vivek Gera, the CEO of Leucine and a seasoned process engineer. Vivek has a wealth of experience leveraging cutting-edge technologies to enhance compliance and operational efficiency within pharmaceutical manufacturing. Throughout this episode, they delve into the transformative impact of AI on manufacturing, QA, and QC processes, shedding light on how these advancements are revolutionizing productivity and adherence to CGMP guidelines.

Vivek’s Journey and Insights

Vivek’s journey into the pharmaceutical industry began in a regulated manufacturing environment, where he gained valuable experience in process engineering and commissioning new plants. His experience across various industries allowed him to draw parallels and identify recurring issues that needed improvement, particularly in the pharmaceutical sector.

Identifying the Need for Improvement in Pharma Manufacturing

During his time in different manufacturing plants, including the pharmaceutical industry, Vivek detected recurring issues that hindered productivity and compliance. He observed a lack of goal-based processes and a disconnect between the management’s vision and the day-to-day operations. This gap led him to recognize the need for a software platform that could reorient systems around delivering critical business goals.

The Solution: Leveraging AI and GenAI

Vivek’s vision led to the creation of Leuzin, an AI software platform designed to optimize manufacturing and quality goals in the pharmaceutical industry. The platform utilizes AI and GenAI to make data-driven decisions, improve cost of quality, and enhance compliance. By activating data from Quality Management Systems (QMS) and Manufacturing Execution Systems (MES), Leuzin enables pharmaceutical companies to make critical business decisions, optimize product robustness, and de-risk themselves against external regulations.

Validation and Explainability of GenAI

In the latest segment of the podcast, Vivek discusses the validation and explainability of GenAI in pharmaceutical manufacturing. He emphasizes the superior explainability of GenAI, providing step-by-step insights into the decision-making process, making it easier to understand and validate the system’s decisions.

The Impact on Quality Assurance Personnel

Vivek also addresses the impact of AI on the quality assurance department, highlighting the potential shift in the need for personnel in pharma. He discusses the potential reduction in the requirement of quality teams and the need for personnel to adapt their skills and knowledge to stay relevant in the industry.

The Future of Pharma Manufacturing and the Role of Technology

Vivek and the host, Yan, explore the future of pharmaceutical manufacturing and the role of technology in shaping the industry. They discuss the agility of enterprises and the potential for smaller companies to leverage AI and technology to compete with larger enterprises.

In conclusion, Vivek’s insights shed light on the transformative impact of AI on pharmaceutical manufacturing, paving the way for enhanced productivity, compliance, and decision-making in the industry.

Episode Chapters:

  1. Introduction to the podcast – 0:00 to 1:20
  2. Introduction of the guest, Vivek Gera – 1:21 to 3:45
  3. Vivek’s journey and founding of Leucine – 3:46 to 10:15
  4. Discussion on recurring deviations and reducing wastage – 10:16 to 17:40
  5. Defining quality metrics and lean manufacturing outcomes in pharma – 17:41 to 25:30
  6. Control and customization of generative AI models for pharma – 25:31 to 32:55
  7. The agility of enterprises and the opportunity for smaller companies – 32:56 to 40:10
  8. Conclusion and contact information for Vivek Vera – 40:11 to 43:10

Podcast transcript:

Please be advised that this is an AI generated transcript and may contain errors.

00:28 – 01:09
Yan Kugel: Welcome to the Qualitaks podcast, which is dedicated to exploring the intricate pharmaceutical industry. Today we are honored to host Vivek Vera, the CEO at the Leuzin and a seasoned process engineer. Vivek has a wealth of experience leveraging cutting-edge technologies to enhance compliance and operational efficiency within the pharmaceutical manufacturing. Throughout this episode, we will delve into the transformative impact of AI on manufacturing, QA and QC processes, shedding light on how these advancements are revolutionizing productivity and adherence to CGMP guidelines. So Vivek, welcome to our show today.

01:09 – 01:15
Vivek Gera: Thank you, Yan, for inviting me. I’m really excited to have this set and looking forward to it. Thank you.

01:15 – 01:52
Yan Kugel: It’s my pleasure, Vivek. So your journey is very interesting. So you’re a process engineer. You worked in various industries before you came to pharma and became also a GSP consultant at some point, and then you decided to found your own company or co-found your own company, Leutin, which is AI software that helps with manufacturing processes. So can you elaborate a bit on this life journey and how it came to be?

01:53 – 02:44
Vivek Gera: Yeah, so Yan, while it may have been different kind of products being manufactured across the kind of companies that I worked. But a common theme would be that I have always worked at regulated manufacturing industries. So there has always been some kind of regulatory body, some kind of regulatory impact that can significantly impact the manufacturing outcome. So I started my journey not in pharma, but in a regulated manufacturing industry, mining. And like you said, I did my education as a process engineer, process control systems engineer, and spent a lot of time applying these skills to the

02:44 – 03:29
Vivek Gera: manufacturing environment. When I say applying these skills, these are skills like being able to analyze the process, identifying opportunities to reduce waste, increase outcome, figure out where are the known value add areas which can be eliminated, identify some what kind of compliance requirements can be streamlined. So I think first of all analyzing the process from improvement standpoint, but then I also spent a significant amount of time commissioning new plants from scratch. So that was the I think part which made me a lot more experienced in how a plant should be designed from day 1 so that

03:29 – 04:01
Vivek Gera: it is set for success and not for failure. So how do you really design the operating system or the control system of how the plant should operate? So I guess, you know, while it has been multiple industries, it has always been a regulated manufacturing industry with the typical manufacturing outcomes at the core, improving quality, improving capacity, improving compliance. So yeah, I think this is a brief description of how the journey has been so far.

04:02 – 04:17
Yan Kugel: And at what point did you land in the pharmaceutical industry and how did you become a consultant in which area did you consult and what did you help your clients improve?

04:18 – 05:11
Vivek Gera: Yeah, no, it was about 5 years ago and I think a pivotal moment came. So in 1 of the manufacturing plants I used to work at, it was a big process plant. And we used to manufacture products which were almost a million dollars per day in revenue. And 1 of those days, the plant was actually shut down for a compliance or a quality reason. And so that was my realization that how compliance can be a very critical function to the sustainability and the success of the company. And continuing on those learnings, I just kind of discovered

05:11 – 05:57
Vivek Gera: that pharmaceutical industry was facing same kind of challenges about 5 years ago And the regulations are a lot more stringent in pharma and rightfully so because we are dealing with human drugs, you know, lives are at stake. So I think, you know, for sure, the idea was to see if those learnings can be applied into pharmaceutical industry as well. So based on that pivotal moment, the second discovery I was able to make quite early on was that not only the problems in the state, but the software ecosystem in pharma was designed in a very peculiar manner.

05:57 – 06:45
Vivek Gera: So most of the software that I was able to discover, they were designed for delivering workflows. They were designed for delivering the day-to-day actions. They were not really optimized for delivering goals, the business critical goals that any manufacturing or any pharma manufacturing facility would like. So they were not designed to, let’s say, ensure 100% compliance with the standards which FDA sets out. They were not truly optimized for reducing the burden of compliance, or in some other words, we can call it cost of quality. They were not really optimized for even measuring cost of poor quality. So

06:45 – 07:09
Vivek Gera: I think a combination of these 2 observations I think truly presented, you know, you’ve seen as a company an opportunity to create some meaningful platform, create a meaningful platform, and go on to the journey. I hope I was able to provide some insights, Yan.

07:10 – 07:53
Yan Kugel: Right, Vivek. And you mentioned that during your time in different manufacturing plants, also in pharma, you detected issues that always reoccur, and nothing has changed in at least 5 years. So can you elaborate on that? It would also be interesting to hear your point of view on moving between industries. Like if you go through, you know, mining industry where you were at and the pharmaceutical industry. So what parallels do we have there that also can be replicated and the knowledge that can be transferred from 1 industry to another and what are the repeating issues that you

07:53 – 07:54
Yan Kugel: can see everywhere?

07:55 – 08:48
Vivek Gera: Right. And I think I was uniquely positioned to draw some learnings out of multiple industries in this case. So I think every industry has its own strengths and weaknesses. What I had done prior to Lucene, I had seen just a lot more automation in the facility than you would typically see in a pharma manufacturing plant. So I think lack of automation was 1 of the weaknesses that I observed in the pharma sector. On the other hand, pharma is really strong in terms of defining what do we exactly want in terms of quality. So I think FDA

08:48 – 09:35
Vivek Gera: has done a brilliant job here, coming up with well-defined guidelines. Sometimes I find that the FDA guidelines can be quite vague. But I also feel that it’s really good because it allows the industry to stay true to science and not really just follow what had been told by FDA. So I think in a way, Pharma does have a really unique strength of being able to define and leverage what is it that we exactly need in terms of quality, so which you can say was kind of lacking in the other industries that I’ve worked. So, unique advantages,

09:35 – 09:49
Vivek Gera: disadvantages, but being able to apply some of these learnings across industries, I think has been 1 of the core strengths that we have always tried to incorporate at least in my work.

09:51 – 10:12
Yan Kugel: And specifically in pharma when you came there you’ve seen the challenges so what did you see that you feel that really needed improvement where you just said okay I want to create this software but I won’t do it in mining or water industry, pharma is the place. So where did you see the need?

10:13 – 10:56
Vivek Gera: Yeah. And Yan, this is where I actually did GXP consulting to begin with. So I didn’t straight away get into solving the problem. I think it was really important for me to get a deep insight into what the problem space looked like. And 1 of the best ways to do anything is to be hands-on. And that’s what I tried to do for almost a year. You know, worked with a few companies as a GXP consultants while I was still learning on the job. But it gave me a firsthand idea of what the current operations, the current

10:56 – 11:51
Vivek Gera: processes to optimize quality, to optimize manufacturing looked like. And this is when it was really clear to me that the biggest gap that needed to be solved was not having goal-based processes. So any manufacturing plant is set up keeping some goals in mind. The goals can be to manufacture as much as possible. The goals can be to optimize revenue. The goal can be to produce the highest possible quality of products. And these goals can be dynamic as well, depending on the market dynamics. And All the work that I did during my education, during my work when

11:51 – 12:36
Vivek Gera: I was working at manufacturing plants, it was always driven by some critical significant goals at the company level. And this was the immediate gap that I was able to observe that the systems in Parma facilities, at least the ones which I saw, were not set up for delivering these goals. And I think it was really obvious to me that if we reorient the systems, the software, the processes, people, tools around single mindedly chasing these goals, then a significant amount of improvement can be made. And when I say improvements manufacturing, you know, let’s say 30 40% more

12:36 – 13:21
Vivek Gera: drugs in the same facility. Reducing deviations by let’s say more than half. Specifically, you talk about recurring deviations, right? So that’s, I think, still a big challenge, as I see. Being able to reduce the wastage, or as I sort of refer to it as the cost of poor quality, where the batches have been rejected due to some unfortunate circumstances or a rework has been done. So avoiding any gaps which come in the way of goals, I think designing your entire system around that was 1 of the obvious opportunities that was evident to me.

13:24 – 13:57
Yan Kugel: Right, so you feel that the pharma manufacturing leaks goals so people just, those are job which so in pharma we have a lot of departments, they all have a common goal, we need to release a batch, but you feel that there is a lack from the management of insight, of vision, of goals that all the employees of the company to work towards or at least in each year plan. So that’s what you mean, right?

13:57 – 14:38
Vivek Gera: The goals are very clear. And that’s like 1 of the another great things that pharma has done. And FD has been kind of supporting that initiative from outside as well. Defining quality metrics. At the end of the day, it’s manufacturing. So defining lean manufacturing outcomes. So there is no ambiguity of any kind in what is it that we are trying to achieve in pharma manufacturing. But the gap is that the way the systems are set up, I felt that they were not very well designed or optimized to deliver those goals. Or if I can elaborate on

14:38 – 15:27
Vivek Gera: this a bit more, we refer to them as business capabilities. So the Business capabilities that any pharma company needs to succeed. So they must have the ability to produce consistent quality drugs. There must be an ability to successfully navigate through audits as and when needed. There should be capabilities around agility as and when some external events happen. Let’s say there is any recall, not for you, but maybe for somebody else in the market. So you might have to respond to it in an agile manner. So, you know, some of the way some of the business capabilities,

15:28 – 15:42
Vivek Gera: you know, I observed were designed, they were not systematically set up to successfully deliver on the manufacturing and quality goals. The goals were, I think, in my opinion, quite clear.

15:44 – 15:49
Yan Kugel: Right. And what was your idea to fix this issue?

15:51 – 15:56
Vivek Gera: Yeah, so I think for me, it was actually quite simple. And

16:01 – 16:02
Yan Kugel: I think this is where the

16:02 – 16:47
Vivek Gera: point which you talked about, cross-functional or experience from other industries comes into play. So most manufacturing industries that I was able to observe in my short span of career, the systems were kind of well-designed to deliver on larger goals, like how much product are we going to manufacture this year? What would be the gross margin? What kind of material requirement we might have throughout the year and then how do we really optimize the cost structure, cost of goods sold. At the same time, because I’ve always operated in a regulated environment, how do we ensure compliance at

16:47 – 17:37
Vivek Gera: all times and then subsequently audit readiness at all times. So I think the idea for me was quite simple to replicate the learnings I had till date, but then contextualize it in the format of pharma and build a software platform ecosystem, which was from day 1 from first principles optimized for goal delivery, rather than doing day to day tasks and jobs and then showing some reports and numbers and dashboards which did not guide any actions. If I have to describe it in a different way build a decision support system. So any good software platform would always

17:39 – 17:59
Vivek Gera: guide the right set of users to make the right decision without them having to really think about it. So building a decision support system was kind of the underlying theme behind what needed to be done to solve this problem and I think that’s what we are progressing towards at this point.

18:01 – 18:06
Yan Kugel: And what stage is the software at the moment? Is it already operational?

18:07 – 18:25
Vivek Gera: Yeah, so we have been operational for about 5 years now. We work with almost 300 GMP sites as of today. So the platform has picked up really well, and very excited to see what the future holds for us.

18:26 – 18:31
Yan Kugel: And does it use AI, as far as I understood from the information?

18:32 – 19:13
Vivek Gera: Yeah, so AI is at the very core. AI has been at the core pretty much from day 1. And the reason we had to do it because, you know, to be able to make any decision, you, you know, need to run various simulations, go through various scenarios, run optimization functions. And I’ll also talk about how even the AI landscape has shifted in the last 1 year and has moved to GenAI. But I think if we are talking about any kind of decision support system, AI has to be at the very core of it. And this entire

19:13 – 20:03
Vivek Gera: foundation has become just 10 times stronger with the additional abilities that generative AI has unlocked. And which is extremely, I think, relevant for pharma, if I may add some more details here. So I think 1 of the unique situations in pharma is that sometimes more than quantitative data, the availability of qualitative data is a lot more. So if I look inside a typical system, Let’s take an example of a QMS, a quality management system. So we have, so in my opinion, QMS is like the central nervous system of the entire manufacturing facility. It’s a place where

20:03 – 20:46
Vivek Gera: you can get a lot of information because it’s like you know it’s a list of all the issues that are happening in the plant. What kind of deviations are going on, you know what corrective preventive actions you took, what are the chain controls that it might have led to, what are the OOS, OOTs which are happening in the facility, are there any issues with the suppliers that we are encountering. So in my opinion, scanning a QMS system unlocks immense amount of opportunity. But if you look at what the foundation is, you will see text documents being

20:46 – 21:40
Vivek Gera: attached in change controls. The investigations or impact assessments are sometimes not available in such a format. They are available in world documents, PDFs. So fundamentally, we have taken the example of a QMS, but I can extend it to other systems as well. The lack of availability of quantitative data, I think has kept some of the possibilities aside, which now can be completely unlocked with generative AI. So I think that layer has improved the foundation by at least 10 times, which is why we have completely transformed the platform, not only to move from AI to GenAI, which

21:40 – 22:31
Vivek Gera: makes a lot more sense given the problem space at hand as well, because GenAI does not only a really good job at handling numbers, but it does a remarkable job at reasoning with the qualitative data that’s available, putting it in the context of science. So not only just reasoning or interpreting the data that’s available, but also evaluating it against the science and theory of pharma manufacturing. What does FDA believe in? So putting things in the context of the science as well is what I think makes GenAI remarkably superior to what we had until a year ago.

22:33 – 22:38
Yan Kugel: And do you implement also GNI in the Lootsin system?

22:39 – 22:44
Vivek Gera: Yeah, so Lootsin platform is today completely a GNI transformation platform.

22:45 – 22:58
Yan Kugel: And what does it help? Does it connect with the QMS of the client or does it connect to the manufacturer and how does it do it?

22:59 – 23:56
Vivek Gera: Right, So since we focus heavily on quality and manufacturing, QMS was the first system we started looking at. So we call it QMS activation. When I say activation, it’s the data activation for making decisions. And like I described earlier, the more we look at QMS, the more I become bullish that it holds a significant amount of opportunity for any company to make extremely critical business decisions. So building upon the theme of QMS activation, it’s now possible to make decisions like, which of my products are robust? And hence, we should continue doubling down on their manufacturing and

23:56 – 24:39
Vivek Gera: their sales and marketing. But on the other hand, which of our products are not as robust as we imagine them to be. And are they the ones that may have to be reorganized in the portfolio? So going all the way till portfolio optimization, starting with very simple things like which deviations are occurring in a recurring manner so that we can measure the effectiveness of Kappa for those deviations and come up with something better so that the manual workload for quality assurance teams can be brought down. And hence, in other words, the cost of quality can be

24:39 – 25:29
Vivek Gera: brought down. So just by activating the QMS data, I am sometimes surprised how remarkable achievement can be made in goals like cost of quality, cost of poor quality, product robustness, and even identifying opportunities for de-risking yourself against the external regulations that might not have been fully assessed on the facility. So we know that recently, nitrosamine regulations came out, right? So a lot of work had to be done and will continue to be done to properly adhere to those. So which all molecules are impacted? What kind of risk assessment should be done? So even assessing and de-risking

25:29 – 26:16
Vivek Gera: your FEI status with respect to the external regulations can be a critical goal that can be achieved by simply scanning QMS. So QMS activation, in my opinion, is an extremely important area. And every pharma company, in my opinion, if not already, will definitely look at in the near future. The same kind of work we have started doing for MES activation, So manufacturing execution system. The problem with that is that the penetration of MES systems is a lot lower than you would imagine it to be. So while there is, I think, an equal amount of opportunity of

26:16 – 26:24
Vivek Gera: MES activation, but it’s quite likely that the lack of penetration might not be very helpful.

26:27 – 27:09
Yan Kugel: That’s interesting indeed to hear about it. So the QMS is much easier to connect to and get our data from. Right. And when you need to validate such a system that makes decisions. So you mentioned that a good AI can help make decision, cut the work time, reduce the cost of quality. So how simple or complex it is and what is the best way to approach the validation of such a system that you need to rely on. So how complex is this?

27:09 – 28:09
Vivek Gera: Yeah, so it is not simple to be, you know, I mean, for sure. However, I want to introduce 1 important area here. So 1 of the unique advantages and unique characteristics of GenAI is that It does a really good job in terms of explainability. So it has, let’s say, looked at your QMS. And perhaps with the combination of continuous regulatory monitoring and being able to leverage the science and technology behind ARMAM manufacturing, it will give you significant decisions. Now 1 of the core, I think, difference between the GenAI and the traditional AI, if I may call

28:09 – 29:03
Vivek Gera: it that, is that the explainability is a lot better. So if you look at machine learning, it was really difficult to sometimes explain how some of the correlations were made. Why should we change the temperature of the vessel to be able to, let’s say, influence a critical quality attribute. So it was more on correlation basis and the explainability was somewhat lacking. But this is what I think is really different about GenAI is that not only are you able to make superior decisions, but it also has much superior explainability. So because it reasons and interprets data just

29:03 – 29:53
Vivek Gera: like a human, using the knowledge that is widely available. It just does it in a much more remarkable manner. But it can give you step-by-step insights into what was the thinking process, what concepts were leveraged, how was the decision made exactly. So in a way, it’s, and I’ll come to the validation in a bit, But in a way, it’s really easy to understand the kind of decisions that JNI would have made. Now, coming to the validation bit. So the way I think, in my opinion, at least the way industries evolving, you know, the technology is not

29:53 – 30:41
Vivek Gera: fully, you know, progressed towards a fully, you know, gam-5 compliant validation ready system as of today. However, it’s moving at a good pace. So the way to imagine this today would be just like you have a really intelligent, experienced person in your facility who is analyzing your systems, who is well aware of the external knowledge, and they are making decisions and they are explaining it to you. And ultimately, a final decision can be taken based on how good the explanation or how impactful the decision might look like. But in a way, it acts like a really

30:41 – 31:26
Vivek Gera: powerful data analyst or assistant or agent in more technical terms. If we invoke some other terminology, we might call it copilot as well. But because it has built-in explainability, It’s really straightforward to use some of the decisions that it can give you. To be able to validate the system, I sometimes question, is it even required or not? But even if it is, I think the technology will progress over time. Right now, everything is moving at a fast pace. So I’m pretty sure that by the end of this year and maybe next quarter, we’ll see some more

31:26 – 31:29
Vivek Gera: case studies coming out on how to validate Yan-EL.

31:31 – 31:56
Yan Kugel: Do you have any case studies where you can showcase how AI has been supporting the GMP compliance and GMP manufacturing, either mentioning your own achievements or also the examples of additional systems that are out there and what is the status at the moment?

31:58 – 32:46
Vivek Gera: Right, so I think this is a really, really interesting topic and there is a lot which can be said and shared at this point. And I’ll give maybe a couple of minor examples here. So there are some opportunities which are pretty straightforward. So it acts as a really good reviewer. So when we talk about manufacturing and quality, review ends up taking a lot of time and energy. So streamlining reviews to the extent that they become nearly automated is 1 capability that Gen.AI has already unlocked for many of the conversations that I am a part of, not

32:46 – 33:35
Vivek Gera: only reviewing, but also generating. And this is how we all are used to by using systems like chat GPT by generating new content, new SOPs, new change controls, etc. So while there are these simple opportunities that I think almost every company has already begun leveraging and taking advantage of, right now the race is towards how do you come up with systems which are not linear, which are not incremental, but they truly become a game changer. So let’s say that no, And I’ll probably not go into a lot of details here, but coming up with new business

33:35 – 34:23
Vivek Gera: models. I think that’s a theme which can truly transform some of the traditional modalities. So I’m not talking about some of the newer ones, but some of the traditional modalities can gain significant advantage by innovating on the business model. And we are seeing some really good examples coming out in the recent times. Simulation is something that I have heard from many leaders. So, if you look at industries outside pharma, then simulation has been a very critical capability. So, before we design any system, we would almost be able to run it in simulation mode, do all kinds

34:23 – 35:12
Vivek Gera: of optimizations, and come up with improvements which are extremely significant, like not 10%, 20%, but sometimes 50%, 60% as well. So being able to perform simulations is something in my opinion that can be quite transformational and I have heard it from many leaders in the industry as well. So while there are some obvious trivial use cases that I think almost everybody is already now doing, the race right now is to create some remarkable and meaningful business capabilities that can take a company from being in the position of competing in day-to-day marketplace to a place where they

35:12 – 35:29
Vivek Gera: almost head towards establishing monopolies. So I think speed will also be a great contributor in achieving that. So yeah, I think a lot can be shared, but I’ll probably take a pause on this point here.

35:30 – 36:21
Yan Kugel: Okay, right. And you mentioned that the generative AI is very helpful here, especially in analyzing reports and creating generating reports. But it means that a person still needs to be involved, right, in the review and the approval of such documents. Of course, it reduces the time to generate such a document because AI can fetch millions of data points, just really quickly analyze them according to the algorithm, according to what you wanted to analyze. But it’s still, because it’s generative AI, it might provide you different results each time you ask for it for an output, right? So

36:21 – 36:35
Yan Kugel: this is 1 of the challenges here to validate because the output is not always the same. And it means that you will still need decision makers to approve each such interaction that makes impact on quality, right?

36:35 – 37:26
Vivek Gera: Yeah. And, Yan, I think you make 2 really good points here. First 1 about the human review and second 1 being about the output of the generative AI being variable across time. Let me talk about the second point first because I think that fits into the review bit. So there are ways to control the output of the generative AI model. And I think this is what will need to be done to be able to truly utilize LLMs as part of your core business foundation. So if you look at any off-the-shelf LLM, they are designed to cater

37:26 – 38:20
Vivek Gera: different kinds of needs. They were not designed specifically for pharma. Even though they do a really, really good job, they, in my opinion, have almost achieved what we wanted to achieve. But the way to truly productionize an LLM for your organization would be to further fine tune it. And I wouldn’t go into a lot of technical details here, but by personalizing, by customizing the LLM to not only a specific domain like pharma, the way we have done it, but for a company who is manufacturing sterile injectable drugs for a specific disease kind of modality, making it

38:20 – 39:17
Vivek Gera: as specific as possible. At the same time, I’ll take a step back. And 1 of the recommended architectures is to create a hybrid LLM architecture where you are relying upon different models for different types of jobs. So for analytics, 1 kind of model may be superior than let’s say another which could be doing a better job for generating SOPs. So I think if you use off-the-shelf generic models, then the output can differ depending on when you ask the question. But we also know why that happens. So it’s because of various technical limitations also that exist. But

39:17 – 40:01
Vivek Gera: it is almost completely solvable by fine-tuning and personalizing the LLMs for the industry and even your specific needs. So that’s 1 of the things we have done. We have taken off-the-shelf LLMs, and we have fine-tuned it for pharma, which took us almost about a year. It was not easy, it was not a very straightforward thing to do. We had to spend lots of efforts, energy, time, even capital in some cases, to truly arrive at a stage where you can start relying upon the output. And which now means that the human review, While in my opinion, it

40:01 – 40:25
Vivek Gera: would still be required, it doesn’t have to be as critical as it would in using off-the-shelf LLM. So I think based on my understanding sitting here today is that while we can design fully autonomous systems because they’re primarily

40:29 – 40:30
Yan Kugel: being designed for decision making, human review

40:30 – 41:20
Vivek Gera: will be required, but the human review will be just to analyze how the decision has been made, what kind of supporting artifacts have been used. It would not really rely upon the ability to generate new ideas or be creative throughout the review process, but to just critically analyze the output. And hence, while I think in 6 months from now, A fully autonomous system can present itself in the pharma context that we are discussing. As of today, some review is required, but it can be minimized by fine tuning your data model so that you are not dealing

41:20 – 41:23
Vivek Gera: with at least unnecessary variability.

41:27 – 41:56
Yan Kugel: Right and what do you foresee for the QA people, right? The quality assurance department. So do you see I replacing a lot of the QA personnel because less review is needed? The report generate themselves, the deviations right maybe themselves through the data that they receive. So how do you see the shift in the need of personnel in pharma?

41:58 – 42:49
Vivek Gera: Yeah, no there will be a shift And we are seeing the shift already. And I can’t disclose a lot of information here. But it would, beyond any doubt, I can say with 100% certainty that the percentage of quality assurance people within any pharma organization can change. And if we take a couple of steps back, I think if we analyze the industry as a whole, the QA team is typically anywhere between 10 to 15 percent of the entire workforce. This is based on some limited analysis that we had done a while ago. But if you look at

42:49 – 43:51
Vivek Gera: other regulated industries, so I’m comparing regulated with regulated, which means that pharma with industries like medical devices, even semiconductors, aviation. So if we compare pharma with other regulated industries, the quality workforce is at least 3 to 5 times higher. And for all the right reasons, because we want to be extremely cautious, we want to do thorough reviews, we want to analyze every checkpoint, because if anything goes wrong, we know that what is at stake. However, with the new capabilities that have been unlocked by the native AI, the same, almost the same outcomes can be achieved by

43:51 – 44:05
Vivek Gera: in augmentation of fewer number of people. So in my mind, at least there is no doubt that the requirement of quality team will go down within this year.

44:09 – 44:52
Yan Kugel: Right, so this is quite an impact on the industry from the quality perspective, right? Because coming from quality myself, I know that there is a lot of manual tasks because not a lot of processes are automated. Like you get a deviation, you need to write everything from scratch. A lot of this can be automated. Finding, summarizing, a lot of those stuff can be then transferred to AI tasks. And in order for people to keep their jobs in a way, what do they need to change in their skills and their knowledge in order to still be relevant

44:53 – 44:55
Yan Kugel: in the industry? What would you say about it?

44:56 – 45:43
Vivek Gera: Yeah, so, you know, I don’t have a lot of qualified opinions in this matter, but you know, I’ll just share what I have thought about this in the past. So I think the situation that we have here today is not unique. Over the last century or so, the technology has kept on improving. And it is just that the technology will continue to improve at a much faster rate going forward. And This is the same we have seen in almost every industry that the rate of improvement has been incremental. So we have not been evolving at the

45:43 – 46:27
Vivek Gera: constant rate, but there is some acceleration in terms of how the technology evolves. And I think we are at a stage now where the technology is evolving really rapidly, beyond comfort. And we started seeing all of this when the cost of computing came down significantly. I don’t have the numbers on the top of my head, but I think it’s somewhere around a couple of magnitudes. So by a factor of 100 or maybe even 1000, that the cost of computing has come down in a very short span of time. So again, I will have to go back

46:27 – 47:18
Vivek Gera: and check the data, but the ability to compute has increased dramatically and which means that the rate of acceleration in the evolution of technology will continue to rise even at this point, which would render some people with skills that may not be relevant anymore. So it’s something that will have to be built as part of the resiliency right from the day when days when people are going through education, even when they’re early in their career. So the rapid learning or rapid shift of skills will become a critical skill In addition to the traditional skills that we

47:18 – 48:09
Vivek Gera: have been having. So there is no, I think, right or wrong answer here, but people who have been working in, let’s say, the quality space, their fundamental skill was not to review the documents. Their fundamental skill was to analyze information and identify anomalies. So that’s how I look at what quality function does at a foundational level. So this need still does exist in the industry. So the need of analyzing information in a meaningful way, identifying opportunities for improvement. So these opportunities are still there. Maybe not so much in the quality function, but any company will need

48:09 – 48:29
Vivek Gera: these foundational skills as to continue to grow. So I do believe that the quality teams will need to start assessing their skill at a much more foundational level and then look at cross-functional evolutions for themselves. I hope that made some kind of sense.

48:30 – 49:11
Yan Kugel: Yeah, yeah, So it made sense. So we see the shift and you mentioned that the technology moves right now very rapidly. So within a couple of years, we’ll see a tremendous shift in automation and AI. So I think that everybody who works in pharma and is used to working in paper and if those people want to stay relevant, they need to be comfortable with technology, with analysis and so on to stay relevant because a lot of the manual tasks will be replaced. So the way for you to stay relevant is to know how to interact with

49:11 – 49:54
Yan Kugel: technology and to take advantage of it to the full extent. And I think that where personnel today will be decided upon which person has the right skill to analyze the data at the end, to sign off of the data, to understand analytics, but also to control the tools and the AI really quickly and rapidly. So you can do 10 times more work, achieve much more work than your colleague who cannot find himself using the technology. So this is from my perspective and how I think it supports what you’re saying.

49:54 – 50:48
Vivek Gera: I fully agree with that, Yan, and let me share another observation quickly here. So when we talk about things like paper, sometimes we automatically assume that people are comfortable working with paper. But in my limited observations, especially the younger employees within pharma, they are all extremely well versed with technology. They are using, you know, softwares at every step of their day, at every step of their life. They’re extremely familiar with all the tips and hacks that go into not only, let’s say, tags, PT, but also some of the other similar systems as well. So I think

50:49 – 51:43
Vivek Gera: even from people’s standpoint, people are, if not already, you know, evolving quite rapidly along with technology. So it’s quite possible that there might not be even any resistance in adopting to the change the way I look at it. It’s just that the enterprises are and can be very slow to move because of just the nature of business. So the agility of an enterprise will, I think, truly determine how quickly the company and the people associated with it evolve into the next era, as I said, transforming business model itself, completely pivoting the capabilities that are required to

51:43 – 52:12
Vivek Gera: be successful in this industry. So I think agility becomes a really fundamental skill and in my opinion the employees or the people working within pharma may already be there. Now it’s just the redetermining step might even be the agility of the enterprise itself. So we’ll see how this pans out, but these are just some of the observations I wanted to quickly add to this discussion.

52:13 – 52:45
Yan Kugel: So that’s a very interesting point because it might be a stage where smaller companies can rise to be giants in a way that they can use the agility and to implement technologies that will help them overrun huge companies, enterprises that it may take them years to implement the same software because they need a lot of approvals and right so that’s a quite interesting view on this. And

52:46 – 53:32
Vivek Gera: again, I think you couldn’t be more spot on at least you know the way I am thinking about this. So finally, I think it is the moment for the mid market, the smaller companies. In Some ways, they got lucky. They really didn’t have to go through all the painful process of extremely costly and extremely inefficient digitalization that the larger companies did. And Now there is a true opportunity for them to leapfrog and straight away jump to the next era. Just like many people never used those landlines, telephones. And I don’t know if there are people who

53:32 – 53:50
Vivek Gera: are leaf-frogging straight away to iPads and not looking at, let’s say, laptops and desktops along the way. So similarly, for the small guy, it is now truly the moment for them to take advantage of the situation here.

53:51 – 54:33
Yan Kugel: Yeah, so that’s extremely a good point, a suggestion to smaller companies. So this is a great opportunity because the huge companies have the advantage of a big size personnel. And now you can cut the cost by using much fewer people to create similar results. That’s quite amazing, I would say. So thank you very much for being a guest on the Qualitex podcast. It was a pleasure talking to you about AI and your solutions. And if somebody wants to reach out to you to inquire more about AI and maybe you’re looking for solutions, what is the best

54:33 – 54:35
Yan Kugel: way to reach out?

54:36 – 55:24
Vivek Gera: Yeah, so people can search me on LinkedIn and from there they’ll find their way to or they can straight away go to and read through what we do. And I would truly appreciate if any of the listeners, viewers have any recommendations for me, any recommendations for the company. So I believe that in all of us are at a stage where collaboration is the key. So I truly look for any feedback recommendations from any viewers to be able to do better in this domain. Thank you.

Tagged as: , , .

Rate it
Previous episode