Validating Clinical Trial Platforms on Databricks

Summary

• Compliance acceleration: Purgo AI uses generative AI with Databricks to automate 21CFR Part 11 compliance validation, reducing validation time from three months to less than a day.
• IQ/OQ automation: The solution automates Installation Qualification (IQ) and Operational Qualification (OQ) script execution, maintaining reproducibility and data access controls.
• Innovation speed: Faster validation enables continuous flow of new clinical trial innovations and implementations on validated Databricks environments.

Validating Clinical Trial Platforms on Databricks

Talk by Kamesh Raghavendra, Chief Product Officer, Purgo AI ; Neha Pande, Senior Solution Architect, Databricks
Clinical Trial Data is undergoing a renaissance with new insights and data sources being added daily. The speed of new innovations and modalities that are found within trials poses an existential dilemma for 21CFR Part 11 compliance. In these validated environments, new components and methods need to be tested for reproducibility and restricted data access. In classical systems, this validation process would often have taken three months or more due to the manual validation process via validation scripts like Installation Qualification (IQ) and Operational Qualification (OQ) scripts. In conjunction with Databricks, Purgo AI has developed a new technology leveraging generative AI to automate the execution of IQ and OQ scripts and has drastically reduced the amount of time for validating Databricks from three months to less than a day. This drastic speedup of validation will enable the continuous flow of new ideas and implementations for clinical trials.
Here’s more to explore:
Unified and open governance for data and AI: https://www.databricks.com/product/unity-catalog
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

Chapters

• 0:00 Introduction
• 0:44 Thank you
• 2:56 Innovation must be faster
• 4:25 Validation is a foundation of trust
• 4:40 Why does validation matter
• 5:38 GCP GMP
• 7:54 CFR Part 11
• 11:33 Demo
• 15:59 Impact on Process
• 18:13 Pergo Architecture
• 22:50 Pergo Results

FAQs

What is 21CFR Part 11 compliance?

21CFR Part 11 is a FDA regulation requiring validated systems for clinical trials; it demands rigorous verification of system components and methods through Installation and Operational Qualification processes.

How does Purgo AI automate validation?

Purgo AI leverages generative AI to automate the execution of IQ and OQ scripts, drastically reducing validation time while maintaining compliance, reproducibility, and restricted data access.

Why is validation speed important for clinical trials?

Faster validation accelerates the continuous flow of innovations and new modalities into clinical trial systems, enabling healthcare organizations to adopt new technologies and methodologies quickly.

How long did validation take before Purgo AI?

Traditional validation processes using manual validation scripts took three months or more; Purgo AI with Databricks reduces this to less than a day.

Full transcript

0:00 Introduction

U thank you. Uh hello everyone. As Alejandra said I want to thank you everyone for joining this early morning session on Tuesday morning. I'm a solution architect at data bricks. I've been working in data bricks for two and a half years now. And today I am joining with hi. Uh I'm the chief product officer of, a data bricks partner and we are eager to present uh a joint solution we have developed together and really appreciate all of you joining us uh early this morning.
You can hop over on the stage. Okay. So before again before we dive deeper I want to take a moment to say thank you. Thank you to scientists, healthcare workers, engineers and everyone who has behind the scene working day in day out tirelessly to develop and deliver vaccines in time in one of the toughest time in the recent history. We if we look back at you know our history we as a human race have have come across many public health crisis but COVID 19 was a wakeup call at a global scale.

0:44 Thank you

It impacted uh in with in every aspect of our life the way we interact to way we develop clinical trials the way we run clinical trials. And now after years of disruption and you know causing the whole life to be slowed down we are finally seeing the life return to a sense of normaly. But with that it also force us to think that how we should think about um speed innovation and responsibility in healthcare. The same co years marked a quantum leap in artificial intelligence technologies with the advent of generative AI.
It opened up a whole new horizon of feasible AI applications in the enterprise. We at AAI, we developed a generative AI technology that is capable of interpreting very high level business requirements that we all have in in a given enterprise and use that reasoning power to generate data products entire production ready data products that can be deployed on data bricks. Um and in this session we're showcasing a joint solution between AAI and data bricks uh focused on the health and life sciences vertical uh that significantly accelerates clinical uh development innovation and productivity and we are very excited to showcase this solution with you and begin partnering with you in this vertical.

2:56 Innovation must be faster

Okay. So when we talk about co or especially the pandemic era there's one key takeaway that we have taken from this is innovation must be faster but it should not be reckless. Meaning we have we have all seen that how rapid innovation and uh acceleration in science and technology can save life. But we also know that one misstep can cause a ripple effect in the entire ecosystem. Especially when we are talking about regulated industry like healthcare and life sciences, right?
So we have to make sure that everything we do is in the boundary of the compliance. So when we when we think about innovation um we all want to do real time OMIX to accelerate our drug discovery. We all want to uh build faster patient cohort with real world integration. um a IML can help in uh disease modeling and patient prediction especially when we are working with scalable cloud native platform but we also know that usually the validation process in a traditional setting can take 3 to 5 weeks that means slow trial onboarding slow submission readiness and then risk of non-compliance without automation so how do we how do we think about all those things and that's Why validation is not just a check box that we have to tick.
It's a foundation of trust that we need to uh have while building a clinical trial platform. So why why does validation matters? Like the question is okay why does validation matters? When we talk about validation, we are talking about data integrity, system reliability, audit readiness. All these things are not just you know check boxes. Again, uh this is what ensure is the data that we are producing today or data we are analyzing today can stand up to the scrutiny in the future be it by regulators, auditors or simply a data scientist who's trying to reproduce the results.

4:40 Why does validation matter

Right? So when we talk about features like versioning, traceability, lineage, lineage tracking, all those features are not just technical add-ons. This this is what this gives us the regulators and the auditors the trust that you know the system we are building is safe and compliant and able to live in the boundary of the compliance. So when we now when we now now now now we have seen how you know validated system should work with the system readiness the audit trails and all those things.

5:38 GCP GMP

But uh when we operate in a life sciences regulator industry like life sciences it especially works under the broad broader umbrella of a compliance called GXP which can be in in terms GCP GLP GMP name it right that's what ensures the product safety patient uh patient safe the data we are using in terms of like patient data is safe and in the secured environment under GXP 21 CFR part 11 compliance has always been a cornerstone of how we manage the electronic records and electronic signatures.
It again it's not just a checkbox. It it is a mandate required by FDA to be a 21 CFR part 11 compliant application especially when you are touching clinical trial data um and clinical research data. Right? So um how do we uh and when we think about all these validation the traditional way of doing this validation have not kept pace with the way uh the technology has been moving the with the fast-paced technology and the complex and sc the scalable cloudnative environments the validation is still slow it it requires you know constant back and forth coordination with multiple teams human interaction and All those steps are like manual which which slow downs the process which slow downs their entire clinical trial on boarding and that's why data bras and pergo have come together to build a solution that would automate that entire process not only accelerating by days or weeks it's it can be done in a minute.
So when we are working with platform like data brakes which is cloud native and changing you know with the change in technology within minutes you can ensure that if something has been broken or if something has been you know impacting the new feature release has impacted your entire process or data that can be validated within minutes. So, AI becomes a gateway between the pre-clinical exploratory kind of stages of R&D, clinical R&D and across the gateway is the clinical stages uh where these types of especially the uh CFR part 11 and other compliance requirements become very stringent.

7:54 CFR Part 11

Um and the way we see it is AI joint solution becoming a a digital backbone for all your platform validation requirements starting with the CFR part 11. Um so what needs is a very high level uh like a natural language based document that describes for your clinical research group what is the standard operating procedure what is the data validation protocol that you'd like to enforce uh built into the joint solution built in the joint solution uh we have a version of the uh installation qualification and operational qualification validation built in as a SOP document within the solution um this is a very simple docu document when you install pergo.
We'll show that in a demo in a moment. You could even edit this document and add or modify requirements based on what your research group uh is complying with. Um so brings the ability to interpret each and every word in your standard operating procedure and associate the intent and the standard operating procedure itself with the underlying data fields which is obviously visible through the unity catalog in your broader databicks data lakeink.
Um this allows uh pergo to use this association and convert your highle natural language based standard operating procedure convert that to a highly granular very specific technical specification. How do these highle legal or compliance requirements? How do they translate? how do they become relevant in your specific set of data fields, specific set of clinical protocols, uh your data pipelines that are processing the clinical data.
Um internally, uses a standard called Girkin. Uh those of you who are familiar with clinical data protocols, you must have used Girkin sometime in the past. So within the product, translates your natural language uh clinical uh validation platform validation requirements to a Girkin standard and further uses this technical specification to actually generate the script the source code of the validation script. Um and this script is then executed across your datab bricks instance all the way from the uh highle pipeline code to the unity catalog to the underlying data bricks instance all the way down to the cloud service platform in which the datab bricks instance is running.
Um this whole execution as I mentioned earlier uh literally takes fewer minutes than I have taken to describe what the product does. Uh and we'll show that in a quick demonstration in a moment. Um so it not only executes the validation script in your platform uh but also creates a a very granular highly detailed uh kind of a a report audit report that can be used for internal reporting external reporting uh file it with your audit colleagues and so on and so forth.
So this entire process which typically used to take 3 to five weeks of manual effort every time there is a change in the data bricks instance or there's a change in the nature of the source of data uh that whole 3 to 5 weeks manual effort is now accelerated to a literal 3 to 5 minutes therefore a two orders of magnitude acceleration in the effort time and complexity of IQoq installation and operational qualification of your clinical data platform.
Um so we'll we'll show here a quick demonstration. So this demonstration runs literally in 2 minutes. Uh but I just want to belabor the fact that in those two minutes you'll kind of see an entire the workflow I was describing earlier uh kind of unfold itself from a high level English language SOP IQoQ platform validation SOP all the way down to running it uh in in a data platform. This is particularly relevant for allowing new upgrades of your data bricks instance uh uh to be instantly validated from an operational qualification standpoint.

11:33 Demo

This allows you to utilize new features especially related to uh machine learning or uh multiomics and the other kind of innovation uh uh items that Nha was describing earlier. uh because of the fact that as and when the data bricks instance upgrades you can immediately validate your platform and your clinical data protocols running on top of that. Um so this is a quick run through of how the product works. Uh so installing pergo in your data bricks instance literally takes seconds.
All you have to do is connect to the datab bricks instance the unity catalog uh to the source code repository whether it's GitHub or GitLab that you're using. So all the scripts generated are stored there and also to the underlying cloud service platform whether it's Azure or or Amazon or any other platform you're using. As you can see as soon as you connect it leaps into action um it it uses the pre-built SOP the text document that describes the IQ converts that to the validation code and starts executing the validation code.
You can see different tests. There are a few dozen tests uh covering both installation qualification and operation qualification. They all run in parallel. Uh and that's how we are able to squeeze all this in a narrow window of 10 to 15 minutes. Um and you can see as the test results come back uh they pop uh the results as as sparse or or fail. You can always drill down, double click, uh compare the result that the platform gave back to with what was expecting.
Uh especially when there is a failure uh there's lot of explainable nodes on why deemed that particular test as a failure. You can further drill down and see what was the real API response uh that the platform gave data bricks platform gave. Um uh the other thing is you can you can schedule this uh in whatever frequency that fits your clinical data uh kind of uh end to-end workflow. It could be weekly, it could be daily, it could be hourly given the fact that how quickly it runs.
Uh you can make it as frequently as you as you want the whole validation. obviously stores all the past validation uh platform validation results. So you can have a a time travel uh like uh kind of go back and forth and see how the platform performed. Um obviously reporting is a key part of compliance. Uh so you can download every validation platform validation as a PDF file. Um and if you further drill down uh the PDF file uh this is very familiar to you.
This is what your manual validation uh teams would have been doing. Um this is now digitally generated. Um it not only has high level pass fail outcomes of individual IQ tests, it also has comments and explanations. Um it also stores all the queries that were generated by in implementing the SOP and the API results that it got. So it's very self-contained um and ready for submitting to any kind of uh internal audit or external scrutiny.
Um, we also integrate this with your overall quality management systems. Uh, IQ and platform validation is one part of end to end clinical data validation. Um, so we'd be happy to work with your QM vendor and your QMS product to integrate with that. We also integrate this with your electronic signature system whether it's DocYsine or any other electronic signature system you use in your enterprise. so that when the quality controller sees the pergo generated report he or she can validate that and digitally sign that uh for submission to compliance and audit.
Um so just to describe the impact of this acceleration uh to your overall uh process uh we'll have Nia walk us through uh some of the actual results uh in terms of uh productivity and acceleration we have seen in customers. Yeah. So this slide tells a compelling story. uh some of which you have first experience in when you validate the clinical applications right um when you're working with traditional SAS software that touch the clinical data usually the IQ setup takes one to two weeks the installation process and then going through the every installation qualification criteria after that usually when these softwares upgrades 6 months one year like the period whatever the frequency the upgrades comes or the new release version comes you again go through the operational testing and uh review of those uh to make sure that you know the release is compatible with the past uh features and nothing is breaking.

15:59 Impact on Process

So that process take usually two to three weeks total in total 3 to five weeks uh every time you know when when the initial setup and then every time two to three weeks additionally when when you go through upgrade and when we think about this this is just the validation but it also encompasses the human error part the coordination with different teams and then constant back and forth that you have to account for. So added more time into it with that you know the manual time takes uh 3 to 5 weeks or probably u more than that when with the automated way the IQ runs not within you know a day but it's it to literally run within a minutes to make sure that you are compliant and the system is compliant with the requirements that you are building.
So the way uh we produce the electronic PDF output shows that even the systems the cloudnative system the modern systems can be as trustworthy as the paperbased documentation that we used to create and providing that electronic record and electronic signature to the auditors to the regulators to ensure that you know we have whatever we are doing the analysis plus submission it's ready and we are always uh the making the system submission I'll take a minute and just walk you through the underlying architecture of So it gives you a sense of how interacts with your data bricks instance.

18:13 Pergo Architecture

Um as I described earlier, is a very general purpose business plus data inferencing engine. It is designed for accepting within your enterprise very high level business or operational requirements and translating the natural language words that your business analysts or clinical associates use translating them in the context of your enterprises own data bricks instance and unity catalog. Um so the the reason why is able to do that is because of the secret sauce underlying in the core product uh which is this graph AI transformer that builds from your Unity catalog and the pipeline code that connects all your Delta Lake uh data fields.
Um so this AI transformer is an organization of all the logic uh all the business and operational logic uh that sits on your data bricks instance all the data fields that are stored in your unity catalog and all the business documents Jira tickets uh AWS or Azure DevOps tickets all the business associate clinical associate business language all these three different sources of data are conjoined uh into a graph representation and further encoded into AAI's language agent.
So this brings a very unprecedented power semantic power that we create for you within your enterprise. It's not like you're exposing your data to any outside service or a external platform because the entire transformer sits inside uh nested inside your databicks instance. does not store any customer data. All data that processes, all data that generates, source code that it generates and requirements it it processes all happen within the confines of your data bricks instance and therefore it gives you complete data sovereignity um and data confidentiality.
Um, pergo as we saw in the demonstration uh not only analyzes and designs technical specifications for high level operational requirements but it it's also has the capability of testing uh every line of code every line of execution uh that pergo generates within the data bricks instance. This is possible because of what I mentioned earlier uh the fact that within pergo it uses the girkin standard uh for generating technical specifications.
Fortunately, Girkin standard is also used for generating test cases. So even before you know the validation script runs, has tested the validation script, proven the assertions that the validation script can really comply with the for example 21 CFR part 11 requirements in IQ. Um so this inbuilt quality assurance uh gives a very high level of predictability and reliability uh that pergo delivers to the customers. uh the end result is a what we call a data product. uh it's it's a complete set of business aggregates uh data pipeline code and validated QA results um that can satisfy the requirements of a said highle you know business requirement or operational requirement like for example the 21 CFR part 11Q requirements uh pergo typically presents this production ready version of the pipeline and data product to the data engineering team for them to inspect and merge that to the production data bricks instance.
So you can imagine now uh moving from preclinical uh data exploration and preclinical R&D studies moving on to more rigorous clinical R&D studies on top of data bricks platform. While you're doing that, you can see plus datab bricks becoming a a new quality assurance, a new data reliability engine and a service that is mirroring the clinical protocols you're onboarding on top of the data bricks instance. Um and given the breadth of compliance that we are building into the product, we believe this validation uh digital backbone will enable uh data bricks life sciences customers to be able to use the power of data bricks and all the features uh that are already there and coming up in the coming months and years for their clinical research as well.
Um this is a snapshot of one of our life sciences customers uh and their own internal um uh uh report of the transformation brought into their operations. U so you can see in the first two rows uh there's clearly a two orders of magnitude acceleration uh in both the reduction of manual effort like validation today is a manual effort that's an example of that but also reduction in time uh to production time to market.

22:50 Pergo Results

Um so you can see literally things that need days or weeks literally become minutes like we saw in the demonstration. There are also other kind of software benefits in using. Um for example because of the inbuilt quality assurance engine and this technical design specification designdriven uh way of execution. uh there is obviously a significant uh improvement in in consistency and code quality uh which is generally the case in adoption of generative AI but doing that at a business requirement level is what is unique with.
Um as you saw any new changes in the requirement any new uh kind of business outcomes that have to be brought in uh can all be pulled in through into the data bricks instance within minutes. So this is a significant boost to the maintainability um of the pipelines uh whether it's clinical or commercial or supply chain related pipelines on the data bricks instance. Um and obviously because this is all done using a a tightly defined set of validations and regulations the level of error and the and the potential for error significantly comes down.
Um again we measure all this very diligently with all our existing users and customers. Um and as we create uh new uh use cases and features for usage of, uh we all monitor this uh kind of tight balance between uh pred predictability and error kind of mitigation uh alongside with innovation and speed and agility. Um so you can see at the bottom the summary results that our customers report consistently showing at least a 90% reduction in time and manual effort uh it takes to bring new use cases online uh on data bricks.
Um so so we are we are here we're very excited about the joint solution for the health and life sciences vertical. Uh both na and I will be here uh even after this session uh through this week. uh we really look forward to having this conversation with all of you and enabling you to use the power and innovation of data bricks as a platform for your clinical protocols. Thank you. Thank you so much.

Learn more about the Databricks Data and AI platform.