Enterprise Financial Crime Detection: A Lakehouse Framework for FATF, Basel III, and BSA Compliance
Summary
- FinCrime lakehouse framework: Databricks lakehouse architecture combines data flexibility with ACID guarantees, essential for financial crime monitoring compliance with FATF, Basel III, and BSA regulations.
- ML-powered anomaly detection: Advanced machine learning models detect suspicious patterns and anomalies in transaction data at enterprise scale.
- Regulatory compliance automation: The framework automates compliance workflows for anti-money laundering, sanctions screening, and transaction monitoring requirements.
Enterprise Financial Crime Detection: A Lakehouse Framework for FATF, Basel III, and BSA Compliance

Talk by Deepak Khetpal, Head of Strategic Growth and BU, Tiger Analytics ; Surya Sai Turaga, Field Engineering, Databricks
We will present a framework for FinCrime detection leveraging Databricks lakehouse architecture specifically how institutions can achieve both data flexibility & ACID transaction guarantees essential for FinCrime monitoring. The framework incorporates advanced ML models for anomaly detection, pattern recognition, and predictive analytics, while maintaining clear data lineage & audit trails required by regulatory bodies. We will also discuss some specific improvements in reduction of false positives, improvement in detection speed, and faster regulatory reporting, delve deep into how the architecture addresses specific FATF recommendations, Basel III risk management requirements, and BSA compliance obligations, particularly in transaction monitoring and SAR. The ability to handle structured and unstructured data while maintaining data quality and governance makes it particularly valuable for large financial institutions dealing with complex, multi-jurisdictional compliance requirements.
Here's more to explore:
Databricks named a leader in the 2024 Gartner® Magic Quadrant™for Cloud DBMS: https://www.databricks.com/resources/analyst-paper/databricks-named-leader-by-gartner
An open, unified approach to your data, BI and AI workloads: https://www.databricks.com/product/databricks-sql
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements
FAQs
Why is a lakehouse architecture important for FinCrime detection?
Financial crime monitoring requires both data flexibility to ingest diverse sources and ACID guarantees for regulatory compliance—the lakehouse uniquely provides both capabilities.
What are FATF, Basel III, and BSA compliance requirements?
FATF (Financial Action Task Force) sets anti-money laundering standards, Basel III covers bank capital requirements and risk management, and BSA (Bank Secrecy Act) requires transaction monitoring—all demand audit trails and data integrity.
How do ML models improve fraud and crime detection?
Machine learning models learn patterns from historical transactions to identify suspicious behaviors, anomalies, and potential money laundering—catching crimes that rule-based systems miss.
Can the framework handle real-time transaction monitoring?
Yes, Databricks enables real-time streaming of transactions into monitored datasets, with ML models scoring transactions immediately to flag suspicious activity for investigation.
Full transcript
All right, thank you. Um, first of all, uh, today we're going to talk about enterprise fin crime detection on the lakehouse framework. Um, quick raise of hands if you're thinking you'll walk off the room and be able to implement the fin crime for your respective finance institutions with the code ready. It's kind of like a sin of sort of way of way of the session. The session is structured where we talk about business, we tie it back to the technology, we switch over to the tech implementation, deep dive on the code, we move back to a business and then tie it back to technology and repeat.
Right. Joined by me, I have a pleasure to um introduce Deepak. Um. I am the field engineering team from the airbricks. Please feel free to meet Deepak and uh he's going to kick off um on the on the on the presentation. Righty. Um thanks. Thank you for sharing the stage. This session was meant to be honestly led by Surya completely. He was kind enough to have me on stage. So why are we here before the drinks, right? So I will do my introductions but that will be the number one question we'll answer.
Then we'll talk about what we have done right so far towards the why and then we'll talk about the how which is what you can take away a little bit with some examples right I agree with Surya it's not going to be a session where you know everything about financial crime compliance and how do you get it rolling right off the bat right so a quick introduction on my side I'm I'm currently leading the financial services practice for Tiger Analytics I've been in the data analytics ecosystem for the last uh 20 plus years worked with several leading global banks extensively in the area of risk management and compliance including financial crime compliance.
We'll talk about what those banks are as well but pleasure to be here. We'll we'll get started. We are it's a small group so we are open to you interrupting at any point raising hands and answering question if a slide has a content which is exciting to you. So happy to hear your thoughts there as well. We will kick it off with the why question. Right? Most of you are aware the why is not Deepak or Surya. The Y is the regulations on the left side.
Right? So this is an area financial crime compliance is the hottest in the banking financial services sector and it's the most global of them all. Right? So there are all kinds of regulations but this one by far is global. So whether you look at the US Patriot Act or the EU AML directive or in Canada or in Australia or even like last year my daughter when she was doing a project with United Nations they were talking about financial crime in Cyprus right teaching them to 11 12 year old kids so it is prevalent so that's the main reason we are here and what are the implications and complexities associated with regulations right it's not going to be a big lecture on regulations here but the complexities are known to all of us right so there is clearly uh a global framework which we have to align on the regulations differ depending on what geography we are talking about the big part is high cost of compliance right so that is now in billions of dollars for some of the large banks it is in billions right so used to be hundreds of millions now it's tracking towards large global banks billions of dollars and the bigger thing is the cost of non-compliance is even bigger right so that's the balance everybody has to deal with so that's the largest complexity we are going to talk about and again um the other part of why we are talking about financial crime compliance in a data analytics and AI forum is because all of this requires lots and lots and lots of data earlier also it was the case but now with AI a lot of AI integration has happened in these processes so we'll talk about that as well and how um you can leverage uh some of the data and AI tools to deal with the regulations so that's the why part first thing we wanted to get out of the way now going a level deeper in financial crime compliance right whatever is the compon component you're looking at of financial crime compliance whether it's customer due diligence or screening or transaction monitoring or investigations right we have kept the flow very high level irrespective of the part you're dealing with there are challenges associated with it right there is huge amount of data as I said but equally important there is lots and lots of manual processes right so If if a bank is under uh a regulatory issue dealing with financial crime, someone like a HSBC or a TD or whoever it might be, they have tens and thousands of people, literally tens and thousands of people doing a lot of these processes manually, right?
They're dealing with data sources in 40 50 different countries from 60 70 sources. They're dealing with knowing for commercial banking, investment banking, they're knowing the entities to let's say knowing the so the right source for each data. All of that is fairly complex when it comes to customer due diligence, right? So knowing different data sources, right source for uh each of the element of financial crime compliance you have to deal with whe when you're doing screening, how do you match the entities you're doing business with the entities which are in sanctions list or screening list.
All of that is again data consuming or manual processes as well. When when it comes to transaction monitoring, there are other challenges, right? You might have all kinds of models to identify suspicious activity, right? So that suspicious activity leads to like tons and thousands of false positive. No matter how good your models are, these activities are so low in their incidence or event rate that you deal with all kinds of false positive.
And again a manual process comes in to decision all those false positive. So there are a lot of these issues in these processes that we have to deal with uh when we we're talking about financial crime. I'll not go into all of it but what is happening rapidly and the good thing for us is with the technology evolution and emerging technologies coming in there are a lot of repeatable processes and automated processes that you can leverage to deal with a lot of these challenges.
So we will pick a few examples of what those capabilities are that you can insert in into your business process to deal with some of these challenges. So that's the context here. With that I'm going to hand it off to Surya to talk about a few of these processes and how we are um handling this. Awesome. Thank you. Um yeah um so thank you for setting the how. Uh quick show of hands of builders and engineers in the room.
Um, awesome, awesome. This slide is dedicated to them where we actually talk about why you're doing the work that you have been asked to do. So, FATF, which is a financial action task force, lists about 40 such recommendations to be done. And what I have highlighted in red um sorry for the blur it's the bigger screen here but um it highlights why you do customer due diligence because it's one of the recommendations that's done by um is asked to be done by the FATF right um the real the real um challenge is there are 40 plus of them and each of these if you get audited by an external auditor you can see on the right hand side of the screen they pretty spend about 18 months digging into your systems, right?
So, they spend significant time within your organization looking into all of this. By the way, I took this from the source is not visible here, but it's actually published by the Tokyo Stock Exchange. Um, so they do a fantastic job listing all of these and some of these slides are actually taken from them. So, now we have the business problem, right? uh what why we are doing customer due diligence because it's a recommendation from FATF once you have that how does it map to different solutions or technology right you can use any kind of a framework to do this and we will slowly map it why it makes sense to do it on a lakehouse framework right uh and then we can throw in or top it up with AI right hey it's 2025 you have not spoken the word AI yet right um so we'll talk about that a little bit but you're doing anti-moneyaundering or you have transaction monitoring systems TMS very popularly known is to do uh and meet certain FATF compliance as processes that you see on the right um so this is how the overall mapping does then comes the BSA right um any any global banks raise show of hands any global banks in the room okay uh so this is very interesting because in the next slide we're also going to talk about an endgame uh which is recently uh being a challenge Um and what does BSA does is it basically uh penalizes the financial institutions for non-compliance.
So please note FATFO has recommendations, BSA is compliance, right? Um so we're not going to get into this, but what you're looking at is u filing suspicious activity report at the time of account opening. Who asks you to do that? Why are you doing that? it's because of the BSA compliance and there are implications when you do not do it right. Um so so one is more about prescription and recommendation the other one is adherence right so we are slowly trying to bring that differentiation about these um banking foundations or financial institution foundations and why it is important as well right um so a typical anti-moneyaundering detection workflow looks like this right and why is it important um for you all to understand this is is powered by a data intelligence platform And what does that mean is you can basically make it extensible.
This is the same framework which will help you not only meet those recommendations but also being compliant right and there are two schools of thought based on jurisdictional or jurisdictional data storage requirements and we'll talk about that in the next slide very very soon. Um show of hands where you run into jurisdictional data requirements day in and out. uh you cannot move this data from this region to that region.
So I see a couple of hands coming off. So that's that's fantastic. Um so this is the idea. You take the intelligence platform, you take the code, you build once you can take the code around not the data. So if you have built a model in a region in a jurisdiction, you can take that any anti-moneyaundering or ML uh with those weights and basically containerize it, dockerize it and deploy it anywhere else and vice versa.
What it powers is um crossarning of the model, right? You cannot move the data around but you can take the model and its weight and move it around different geographies so that it learns from each other. Right? This is how you build a robust system. An example, when you travel international, you probably have to call up your respective credit cards or institutions saying that please allow this transaction or they ask you for the first crossjurisdictional transaction, right?
But the fraud or the anti-moneyaundering that happens or the fin crime that happens does not understand jurisdiction. Your code and model should need to understand that, right? Um so that is a critical difference. Now just double clicking on one of the others which is the three pillars um which is pretty much developed by the committee uh and the BCBS team and it has very important three um requirements that we stress about and I'm there are so many but I'm stressing only on the three because it it piggybacks on what is called as the 3 endgame.
Have you guys heard about the 3 endgame? Fantastic. Um and that actually talks it's a very interesting schools of thought that comes into the picture where whether the financial institutions are powered enough to support SMBs or small or medium-sized businesses to issue or grant a loan versus the capital requirement and its implications for all of these financial institutions once the loan is approved. That is the math that basically goes into each of them.
Um so there is an EU regulation where I think Tiger Analytics has a well curated solution around it. Um we'll talk about that deep will stress on it uh in the few slides but what you're looking at pretty much is the standard calculation of risk weighted assets um the total of tier one tier two what's the buffer that needs to be there uh for gibs um and so on and so forth right um so these are standard but why we stressed upon it is um the basil 3 endame have you guys seen the movie bank of all right so u this is inspired this slide is inspired by the movie called as Bank of.
That basically covers your 3 endgame in that hour and a half movie. So if you get some time, you want to relax after the long day, do watch it. Uh I did not use the word bank of, I said bank of brave. Um but yeah, that's that's essentially highlights key aspects about the 3 end game and why the bigger financial institutions pretty much um cannot approve these loans. uh and there are requirements of these small banks where they don't have the need to have the securities listed and submitted uh to have the valuation and then approve on the capital requirements and so on and so forth.
Awesome. Now let's get into so that's the business context, right? Let's get into the solution and the technology. The we're going to run through these slides really fast. We're happy to move off stage and answer some of these questions, but we're going to talk about the seven pillars. Uh hope time permits and we get into some actual implementation and deployment considerations. We do have the demos as well. Um but I think we don't have the audio video playback support yet.
But so we'll talk about that. Happy to show you the demo as well on the database platform and how it works. Um let's talk about pillar one. So if you look and remember the data intelligence platform for the AML um this is the data bricks data intelligence platform for the entire stack. Notice the things that we have in yellow. Um those are the things with the data bricks as a capability powered by AI. Now if you just look at suspicious activity report it is by regulation that you cannot use automation systems to submit the S.
It has to be a human who needs to look at it. However, you can power that human behind the screens with the technology. That's the rule, right? So, why we are talking about this is it's almost like your tax submissions. Uh it cannot happen automatically. There has to be somebody who's authorizing and digitally signing your tax and you submit it. It's very similar to that. Um so why we talk about that is um the critical aspect of it is the first one the data layer deep pretty much mentioned about how you have 70 to 80 systems coming all together but what I am stressing here and you probably have seen this a gazillion number of times all throughout the day but what I'm highlighting it what actually goes in every layer of the medallion architecture right so if you are ready to start off this project what should they be there in the bronze layer what should happen in the silver layer and what should happen in the gold layer as a recommendation or a best practice.
Right? What sits in the bronze layer and for what reasons uh is mostly about governance, the lineage. Um and it'll all tie up together uh in the next slide where you will actually see how you will implement some of these at the bronze layer and what schema definitions what are the core data model implementations that has to get go in and what do you recognize? How do you maintain quality at the silver layer and finally publish something as call as a customer risk profile as a data product in the gold layer.
Right? So this is a sample example um of how you could you could put together. Now since we have captured what goes where in the medallion architecture now let's look into the processing component. Right? So we looked at customer risk profile but let's do a double click on what does a risk data system looks like? What does it encompass? Right. Um so you have all the different types of data sources. Um the operational systems can anybody take a guess what are typical your operational systems sitting today?
Most of them it's still mainframes believe it or not right nobody has dared to touch the accounting systems that's running for centuries if not decades. Decades is still old. Um so so yeah so you that's there that's where it allows you to create what we call it as the authoritative data source. Right. So the creation curation of an authorative data source is super important and that's where the fundamentals of a risk data system when before even you publish a customer risk profile starts to starts to evolve right then questions about federation versus ingestion come around.
What are those technology tools stack? How do you bring the data to connect to your on-pre systems or mainframe systems and ingest that data into the cloud um and do rest of those data processing at scale and so on and so forth. External data source is also getting into what is called as a central data roy. I call it a thod data data source. But the part of the curation is just that little cylinder at the bit in between.
The rest of the components is where you will actually spend the most of the time and those blocks sizes actually highlight what goes behind these. Right. Um can you tell me in the last five years what is the most popular external data source for a financial institution? SAP watch list. Awesome. Yeah. What else? Done at Brad Street. Okay. Yeah. Interestingly, um, how many of you have actually done um, return of the goods that you have purchased as no longer needed?
That's an interesting external data source. I want you to spend 30 seconds thinking about the example that I've coded. How that is implying on crime. It's a huge area where all of this are actually coming together. The data sources, external data sources are very very getting very very interesting very fast. The future however is going to be agentic systems. Exactly. Right. Um so give it a thought and why this diagram is is important.
Now and hence comes the term about unification. Right. unification becomes reality suddenly very soon very fast because you are combining two or more than two types of architecture patterns. Um and this is why you are actually able to meet more than those 40 plus recommendations within the same architecture. So unified architecture is the key word not unified data. Good luck building a lake where you're trying to bring all of your data at one place. um and then try to build a system around it.
It's almost impossible. It's impractical. But you can have an architecture that powers that, right? Um so that's the key. Now just looking into the compliance implementation part, we looked at FATF, we looked at 3, some of these BSA compliances and so on and so forth. Um do you guys know you can actually automate the process of validation of these STR reports? So whenever this goes to finense they provide a schema definition the XML XSD schema definition for you to validate whether whatever you are doing as a submission for S is actually right or not right that capability good luck building that on prem right because you just don't do it once and you're right at the first time you iterate over it right very similar to your tax submissions how many of you right the first time after doing it for decades that yes I'm confident that I made the submission right almost less than 50%.
Right? So you do this exercise that means repetition extensibil extensibility are key for your architecture and all these compliance implementations. I have my own GitHub link outside. Uh we would like to we could go to search for on my name and you will see the entire implementation of it. I'll probably give you a glance of it. Um and also the Python XSD where you can validate which is published by Fininsson on the S activity report when you submit it to the Finson's.
Right. So this is all available for you. Uh so that's the QA side of things. Now the most important thing which ties back to slide number five that I was talking about where an external auditor comes in and audits you for 18 months right that's a typical timeline do you know why a majority of the time is spent it's spent in giving that auditor access right I can I can see smiling faces a very practical reason and that's where the unity catalog and the governance comes into the picture um where we talk about all of this.
Now guess what? If the auditor finds out that you do not have enough data protection or fine grain access control or the ability to do lineage, right? Um so this is what data bricks brings onto the table. We talked about Unity catalog and governance, but I'm trying back to the practical reasons why this is important for you as a financial institution to implement. Right. Um now I want to pivot back a little to debug and what tiger analytics have um a readym made recipe uh about mlops and model governance.
But before we go into this topic, I'm going to just put a thought on the previous part. Right? The data aggregation part has been the single largest reasons why most of the penalties and fines have been levied. Right? The auditors, the regulators, they're looking for your processes of submission. If you're not able to explain how you got this loan number or this submission or this activity and map it back to the complete process then there is trouble right and depending on what is the category of bank you are whether you are a GI or a category 1 2 3 4 the regulatory requirements obviously differ the heightened standards come come into play I'll not go into any of that or all of that but obviously that's the reason this is so important.
What we're going to touch upon in the next five seven minutes very quickly is an emerging area of again heightened regulatory compliance requirements which is model risk management or now with AI it's also AI risk governance right so the fundamental pillars of model risk governance are still the same as they were defined almost 14 15 years back right which is your SR R 117C guidelines, right? You want to be able to validate the input data that you've used to build your models, whether it's your financial crime compliance models, AML models, whatever suspicious activity models, the input data that was used to develop the models to the approach that was used to build the models to the next step of validation of those models to see if they're performing well or uh they're performing uh as expected.
Right? So depending on the type of model there are different types of metrics you're supposed to monitor uh whether it's sensitivity stability all of that. So that's the third area right the validation of it then before you are doing the implementation the testing of it. So there is a whole framework around it right. So that has not changed. What is changing rapidly is the number of models that are coming into play and also the types of models they are changing completely meaning five 10 years back all you would hear about is a logistic regression model to whether it's detecting fraud or cross-ell or upsell to acquisition all of those were simple models now we are talking about AI models vendor models they are complete black boxes even the developer of those models don't know what they have built when you come when it comes to LLM models the hallucination and all of that hence the model governance part of it has become all the more important so increasing number of models increasing complexity of models and the third thing is your ability to even measure a lot of these things on the left right the conceptual soundness meaning how do you trust the model is very hard a AI model all kinds of firms are struggling to figure that out right so that's why it's it's again becoming very important and relevant and it's a hard a very hard problem to solve right so that's uh that's the reason we are going to spend a little time on on that and with AI governance same principles apply it's just that it applies on the AI models right but AI models also have existed for 10 plus years, right?
Genai models might not have existed for 10 plus years, but AI models have uh existed as well. So that's the context of model risk management and AI risk governance. The second piece we going to talk about is meaning I covered some of it. Um right and I think now the right side of it is also becoming a lot more important right the transparency and explanability it is all the more important in banking financial services where bias is an issue or uh uh meaning again on AI governance you have heard a lot of this as well so that's again the business challenge uh what we have done in collaboration with data bricks and uh obviously with some of our clients is a solution which is completely built on data bricks uh environment.
It's a brick builder solution now uh which has been deployed with some of our leading clients again leading financial services firms as well as some non-financial services institutions as well which does all of those five or six things we just talked about in a automated framework right so when I talk about automation right from your ability to ingest the data from again the data bricks architecture or whatever you might have in your environment bringing it to the data brick um infrastructure let's say bringing those models into that infrastructure to bringing the data into that infrastructure all of that has been built on data bricks environment including let's say the observability of those models right so uh we meaning whatever metrics you are measuring all of that and what we have also done is codified what we know typically happens in financial services space and not at the highest level but depending on the type of model.
So there are close to in a bank even retail bank there are going to be like literally 150 odd models for a midsize bank right fraud risk models credit risk models uh AML models so depending on the type of the models the metrics that are needed to measure might be different so we have codified a lot of this into the platform as such which is on data bricks now um as well um I'm I'm I'm going to not go into the detailed technical side of it.
But again, if you look at the architecture, it does cover a lot of those circles that we talked about. Uh and each box of it is and by the way, even for this, we have a demo available. Um uh you can join us at Tiger Booth. But um for each box, we have a component which addresses that right again built on data bricks. uh whether it's the governance side, whether it's uh the workflows on how do you route these models, whether you uh designing the security systems around it, all of that is uh available um on data bricks platform as well.
Um I'll come back to you uh Surya if you want to cover. Yeah. Yeah. So um thank you. So that's that's about like a specific solution overview and implementations on what could be covered now. Once you have implemented uh once you have auditors you've given them access to data and everything else what else they are interested is pretty much about the moderate auditability and explanability that Deepak was talking about right um so it's very important that you and your organization start looking towards Python based models very fast um do the sensitivity overlay analysis um that is required but the most important bit is model auditability and when when you are when you're asked what they're looking looking at specifically is do you have an AB testing framework right or how do you do AB testing how do you launch these new models or many more models as in blue green deployment um and how does you how does a user adoption looks like and why does the a particular model was used what was the performance of these models yet you can have models with large weights but what's the point if it takes a long long time to detect fin crime over over a a moving moving data stream right so so all of these um very interesting and this is the area where I'm actually uh looking deep is what is called as regulatory documentation right pick any financial institution the biggest challenge they have today is auto documentation right and this is where a lot of promises from AI has been made uh but the reality is the format the structure in it needs to be done is very fixed so model um and and AI services can curate or create something that doesn't exist.
That's the biggest risk um and exposure to the bank or a financial institution. So um this is where you could do um um and and I'm I'm looking at certain uh products out there in the market. Uh some of them like valid minds and all uh they're doing pretty good job but data bricks believes in um an open framework like MLflow. So if you are creating some things around or innovating in your own streams about auto documentation uh do take a look at MLflow and their APIs the faster you can integrate the whole ecosystem can completely be documented automatically and it'll be more and more deterministic than probabilistic.
Um so I would encourage everybody to look at open source ML flow um and and tie it back to those APIs that is already available. Awesome. So these are some of the examples how um I had to I have to blur some of those but you can see um how frequently you can do some transaction monitoring uh what's the precision and recall rate looks like what's the frequency they are interested in how many trials you have done uh before you launch a particular u model into production and so on and so forth.
So those are key metrices that um are required for you. Now uh specifically talking about performance monitor per performance monitoring what exactly do you need to monitor right from a business and solution standpoint what are those KPIs that you should actually be building a framework around monitoring right and how frequently you should monitor it right so some of things that I've listed here is like a daily daily frequency um and what kind of alerts you need to put into your system when you're building this whole um architecture on data bricks and data intelligence platform Um there are some mandatory ones which you will see some compliance metrics right what are the filing timelines look like when was this last submission done what date and so on and so forth what is the quality score of your data when you are when you're what is what is the data certification process looks like those are some of the things um that you should be actually monitoring as part of your KPIs um of course on the RWA side of things risk accuracy and so on and so forth now guess what the the solution doesn't end there, right?
Uh what we are looking at pretty much is what I started talking about where your you know operational systems are and what they are. You're looking at some things what we call it as um extensibility of the solution that you're building together. You should always make sure that your solution is open source not blackbox. When we say open source that means there is open APIs to be able to talk to them and external systems that you could push this data out because the reality of the matter is wherever you build you probably need interoperability between your on-prem systems and the cloud.
So the systems that you're using or the APIs that you're using should be very much open source friendly to make that happen. Awesome. Um so while based on our experience at doing that at uh some of these financial institutions I'm trying to share some of these deployment considerations. Um so um something that I want to quickly touch base about is um some things which is mandatory is your resilience and disaster recovery.
Um so so the systems that you build could be active active active passive based on the system and the compliance. Um the last thing we want is you're getting audited and your systems are down, right? And you do not have a BR BCPDR play in in place. Uh so that's the last thing you want to avoid. Um and the other thing I also am especially interested in was on the CI/CD process. You did the model development in the cloud, you have the weight, now you want to retrain it um or or tune it with the data that is available on prem.
So you cannot take the data on cloud but you can take the model and the weight to the on-prem and again retune that right. So there has to be a continuous CI/CD process. We look CI/CD process from an aspect of very much doing end to end on the cloud or on prem. But what I am looking at is going between on-rem and the cloud right so the compute exists somewhere else the data exists somewhere else but you can take the code and the model and its weights and try to tune it um in your own on-prem systems and so on and so forth.
So um some of these things I would say are key but what I'm highlighting in this slide is don't just think about the data and the codes recovery but now also start looking about the metadata replication because without that how would you traverse the lineage part of it what is the original system what was the operational system and what was the external data source all of the metadata also needs to be replicated on prem for jurisdictional requirements all All right.
So that's where the key um aspects are. With that I um we are almost at 5 minutes. Um so concluding this session. Uh any final thoughts Deepak? Just one thought right we covered a lot of ground from highest level of regulations to certain challenges in financial crime to some repeatable components. Right? But none of this applies to any bank or any financial institution as is. Right? This is just frameworks and some in some cases repeatable frameworks and repeatable templates all open source.
All of this has to be customized to your own situation, your own challenge. And I I think the beauty of it what Surya also mentioned is you you can have a starting point versus having to build from scratch which is a much difficult problem to solve. And also there is something which which has proven to work with some of your other peers and hence you can feel little bit comfortable repeating it for your own bank quoting to the regulators that hey here's what some of my peers are doing we are we are replicating some of it uh to ensure compliance with that we'll take a pause any questions happy to answer I know we have five minutes now but even after that over a beer over a drink uh we are happy to continue the conversation
Learn more about the Databricks Data and AI platform.
The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.