About
I've spent the last decade alongside fraud and financial crime teams at some of the…
Articles by Karthik
Activity
2K followers
-
Karthik Tadinada shared thisBeing taught a lesson in patience and resilience on the way to the FRC Leader’s Convention in Newport. 0648 : Delayed -> Cancelled 0712 : Cancelled 0748 : Delayed -> ?
-
Karthik Tadinada shared thisSuper excited to be at the Cambridge Claude Code meet up on the day Anthropic released Opus 4.7. Thanks to Redgate Software for sponsoring and the Bradfield Centre for hosting. They even managed to arrange for good weather!
-
Karthik Tadinada posted thisThe SaaS model is quietly breaking down in fraud and AML. If you zoom out, the progression of software delivery has looked something like: 1. software shipped on physical media 2. on-prem installations 3. cloud software Each step in the wave delivered more value to the customer and did more of the job. Despite what you're currently being sold, the next stage isn't just legacy SaaS with an AI layer slapped on top. Bolting new tech onto old architecture doesn't deliver the actual benefits of AI. It just creates a more complex tool – and drops the problem right back in your lap. You still have to become an expert to get any value out of the technology. The coming wave is solutions – providers becoming genuinely accountable for the outcomes they claim to deliver. Bundling software, maintenance, inference, implementation, and financial crime expertise into a single offering. Staying in the room when it gets hard. Because the real result – reduced fraud losses, lower false positives, stronger AML coverage – comes from the full stack of implementation, strategy, tuning, and expertise. Not from a software licence and a login. That's why at Fortify, we made a different choice. The outcome is the product. We don't unbundle expertise from what we sell, because that breaks accountability. Our job is not to ship software and step away. Our job is to help customers solve their problem. The companies that define this era in fraud and AML will not be the ones that made the most noise about AI. They will be the ones that built genuine trust with the people doing the work and solved real problems. Are you tired of buying "solutions" that turn into second jobs for your team?
-
Karthik Tadinada reposted thisKarthik Tadinada reposted thisI wanted to take a moment to thank the Founding Class of practitioners who believed in Safeguard early. Many of you committed before the agenda was finalized. Before the speaker list was announced. That says a lot. It means a lot to see people back something because they believe the industry needs it. Thank you for being part of the founding class and helping shape what Safeguard becomes. TODAY IS THE LAST DAY TO REGISTER! We’re about 12 hours away from closing registration. Registered attendees have already started building their experience. Profiles are open. Opting into roundtables. Choosing the experiences they want to be part of. The best conversations are already being locked in. Qualified practitioners receive a complimentary pass and $1,500 travel reimbursement. Save your spot here: https://lnkd.in/eA2STbUY If you've been on the fence… I'd love to help get you there
-
Karthik Tadinada shared thisI watched two investigators handle the same alert recently. Both had the same alert, the same data, the same rule. The first saw a construction business spending big and moved on. The second flagged it. New merchant, unusually busy, then nothing. That didn't sit right. The second investigator was right. It was fraud. The first investigator made a perfectly reasonable call. They just didn't know what the rule was trying to catch. Nobody had told them to look at merchant age. This is one of the quietest, costliest problems in financial crime operations. Not that alerts don't fire. They do. But the alerts don't tell investigators what they're looking for. So outcomes depend on who picks it up, what they've seen before, what their instinct says. Flagged fraud gets through – not because of bad people, but because of missing context. There's a straightforward fix. And it's the kind of thing that once you see it working, you wonder why you weren't doing it already. That's Lesson 4 in our Field Notes series – 8 lessons drawn from two years embedded inside banks, fintechs, and payment processors. Comment below and I'll send you the lesson!
-
Karthik Tadinada posted thisWe often talk about a leader's "gut instinct" as if it's some magical, innate gift. It's not. I recently heard a speaker challenge the way most people think about intuition – and it completely reframed how I think about leadership development. The argument was simple but powerful: intuition is really just accumulated decision-making experience. It's what happens when you've been exposed to thousands of decisions, received fast feedback, and operated under pressure – over and over again. What struck me most were the examples of how elite systems engineer this on purpose. The military doesn't wait for combat to develop leaders. They create compressed, high-frequency decision environments in training – low stakes, high reps. Football academies do the same. Barcelona's La Masia uses smaller pitches and smaller balls. Bayern Munich trains with tiny goals in impossibly tight spaces. Why? To massively increase: - The number of touches - The number of decisions - The speed of learning The principle is the same: you don't wait for real-world pressure to build judgement. You manufacture it. There's a lesson here for how we develop leaders in business too. Strong leadership and sharp judgement aren't some vague instinct people are born with. They're built – through repeated, intense decision-making. And this is where Gen AI becomes genuinely powerful – not for the reasons most people talk about. Historically, so many business decisions required huge amounts of tedious data preparation and busy work before there was even enough information to act. That friction meant fewer decisions, slower feedback, and fewer reps for your people. Gen AI removes that friction. It compresses the time from question to insight, which means your teams can now make and monitor 5-10x more decisions than before. It's the business equivalent of Barcelona's smaller pitches – dramatically increasing the number of reps. Done well, this should change the trajectory of your company. Not just because you're operating faster – but because every one of those additional decisions is a rep. More reps, faster feedback, higher frequency. That's how you build leaders with sharper intuition – the same way the military and the best football academies do it. The question for any organisation: are you creating enough reps for your people?
-
Karthik Tadinada shared thisIt is really exciting to hear other people hop on the “warehouse native” train. This is the bet we’re making at Fortify Solutions | Financial Crime Prevention together with our partners at SnowflakeKarthik Tadinada shared this"A warehouse like Snowflake/ Databricks/ Azure Data Fabric. That’s where most of the data now lives." "Once the data is there, companies are building context graphs and governance layers on top of it to feed their agents. Microsoft Copilot is a strong example of this direction." "We are technically in the third year of the agent cycle. Agents really started to take shape in early 2024, so by 2026 we’re already fairly far along. In sales and GTM, it’s still early, but more broadly, agents have been around for about two and a half years." "People tried to build agents on top of legacy systems, and they didn’t work because the data wasn’t there. So what’s happening is exactly what Satya Nadella has been pointing to: businesses, especially in the Global 2000, are centralizing their data into their own data fabric." "Rox is warehouse-native, meaning we run inside the customer’s environment on top of the data warehouse, and it dramatically improves time to outcome because the context is already there." Ishan Mukherjee (Rox) on the Agentic OS Summit with Dina Habib OMara (Microsoft)
-
Karthik Tadinada posted thisWhen Amazon launched, the story people told was about books. But the real shift had nothing to do with books. About twenty years ago I became genuinely fascinated by ferrohydrodynamics. If you do not know what that is, you are in good company – it is a niche corner of fluid mechanics involving magnetic liquids, and the number of people who want to read about it in depth is not large. Before Amazon, that was basically your problem. Publishers did not print for audiences of a few hundred. Retailers did not stock for curiosity that specific. The market had decided your interest was too small to serve. Amazon changed that. Suddenly every specific, niche, previously unserved interest had a solution. It did not matter how obscure the problem was. If it existed, you could now get help with it. I think GenAI is doing exactly the same thing for fraud and AML – and most people have not quite registered it yet. For decades, fraud teams have had to work with software built for the average customer. Every vendor makes the same calculation: what is the most common problem across the most customers? Build for that. Everything else is a feature request that goes on a roadmap and probably never ships. The result is tools that are fine for generic use cases and genuinely poor for specific ones. Your customer base is not average. Your geography is not average. Your product mix, your risk appetite, your regulatory environment – none of it is average. But the software does not know that, and it was never really designed to care. GenAI changes the calculation entirely. For the first time, you do not need a vendor to have decided your problem is worth solving. You can describe your specific typology, your specific data, your specific context – and get something genuinely useful back. Every specific, previously underserved fraud problem is suddenly within reach in a way it never was before. But only if your data is accessible. This is why the architecture conversation is not a distraction from the AI conversation. It is the AI conversation. If your data is locked in a proprietary system that does not expose it in open formats, you cannot get there. You are back to whatever your vendor has chosen to build for the average customer. The GenAI era arrives for everyone else, but not for you. The fraud and AML teams that are going to get the most out of the next few years are not necessarily the ones with the biggest budgets. They are the ones who made their data accessible early – who can point an AI at their specific problem and actually get a useful answer. Amazon did not win because it had better books. It won because it made more problems solvable. That is exactly what is now available to fraud teams. The question is whether your data foundation lets you reach it.
-
Karthik Tadinada posted thisThe Claude Code source code leaked yesterday, 512,000 lines, 1,900 files. And the internet is going wild. But here’s the part I think most people are missing. Strip away the complexity, and what you're looking at is a very sophisticated system for one thing: automatically constructing the right prompt so a model can generate good code. Every tool, every slash command, every memory layer – it's all scaffolding that spares you from the nearly impossible task of writing the perfect instruction yourself. What genuinely stopped me in my tracks is how familiar the core loop looks. At a high level, it resembles the same basic agentic pattern you can see in mini-swe-agent – a roughly 100-line agent that is already competitive on SWE-bench, showing just how far a very small loop can go. What stands out is how much product infrastructure sits around that basic loop. That's not a criticism. That's the product insight. The magic was never the model. The magic was knowing exactly what context, tools, and orchestration the model needs to do a specific job well. Which brings me to what we're building. At Fortify, we're applying this same principle to fraud and AML strategy teams. These teams design detection rules and risk strategies daily – and the gap between a general-purpose AI tool and one that deeply understands their domain, their data, and their job to be done is enormous. We're building the "Claude Code" for fraud and AML: an agent that encodes domain knowledge so precisely that it outperforms general-purpose tools on accuracy, latency, and cost. The lesson from the leak isn't that Claude Code is simple. It's that deep specialisation compounds. A 100-line loop becomes a category-defining product when it's built around the right domain. More on what we're building soon.
Karthik Tadinada replied to a comment
1d
Oonagh van den Berg it’s my first time here and it’s a great event! I think that dealing with delayed/cancelled trains on top of long absences would tip anyone over the edge!
Karthik Tadinada commented on a post
1d
Update : happy news! The 0748 left roughly on time! Hugely looking forward to the conference!!
Karthik Tadinada replied to a comment
4d
Timothy Van Heest that’s where my interest started too, auto suspensions! Such a shame that they ended up being very power hungry
-
Karthik Tadinada reacted on thisKarthik Tadinada reacted on thisI am incredibly proud to announce that Amotivv Inc. has successfully closed our first angel investment round! 🎉 As Founding Principal and Chief of Strategy, I know that ambitious visions require bold partners. I want to extend my deepest gratitude to our six incredible angel investors spanning the US, the UK, and Italy. Your early confidence in our mission to bring transparency and auditability to AI is the catalyst we need to scale our solutions globally. Thank you for partnering with us on this journey. 🌍🤝 Laurence Garrett, you were exceptional guiding us on the board at Featurespace, and I cannot imagine a better person to join us as our first angel investor. Kevin K. and Geordie S. we’re similarly thrilled to have you joining us as angels, where your wisdom and experience will be a tremendous asset paired with our innovation and momentum. Adding to the momentum of this close, I am thrilled to officially welcome Kevin Shamoun of Fortispay to our Advisory Board. Having his strategic insights and industry expertise in our corner will be invaluable as we navigate our next phase of growth. Executing a successful funding round requires an elite support system. A massive thank you to our legal partners at Wilson Sonsini Goodrich & Rosati, the preeminent software and AI law firm in the world. Special thanks to Bradley Doline , along with Daniel Glazer and Marty Korman from the London and M&A practice, for their world-class guidance, dedication, and counsel throughout this process and my career. ⚖️ Now, the real work accelerates. To deliver on our promises, we practice exactly what we preach. Alongside our human team, we utilize five AI personas powered by our own Synthesis product. These AI assistants work alongside us in critical daily roles, allowing our lean startup to punch far above our weight class and proving firsthand the power of governed, human-AI collaboration. 🤖🚀 Backed by global investors, guided by top-tier advisors, and powered by a hybrid workforce of humans and AI, Amotivv is primed to build the trust layer for the future of artificial intelligence. We are just getting started. Check out what we’re building at www.amotivv.com! #AngelInvestment #FundingRound #AIGovernance #ResponsibleAI #TechStartups #Amotivv #ChiefOfStrategy #FutureOfWork #AIIntegration Russell Moore Jason Smithamotivv — Enterprise AI you can trust, audit, and keep.amotivv — Enterprise AI you can trust, audit, and keep.
-
Karthik Tadinada liked thisKarthik Tadinada liked thisSnowflake Cortex Code (COCO) ❄️ might be one of the most practical AI releases for data engineers in a long time. Not because it sounds impressive. Because it actually helps me ship better work inside Snowflake. I’ve been using it for testing, fixing code, and designing better ETLs, and it’s starting to feel less like an AI feature and more like a practical part of my workflow. My personal view is simple: Data engineers do not just need code generation. We need context-aware assistance inside the environment where real work happens. That is why COCO stands out to me. There are many code assistants in the market today. But in most enterprises, adoption is still not fully there when it comes to terminal-heavy tools like Claude Code. Not because they are not powerful. But because enterprise teams still care deeply about security, access control, governance, workflow fit, and platform trust. 🔐 That is exactly the gap COCO is filling and filling it very fast..... Because it works where many data engineers already build: inside Snowflake. Here’s how I’ve been using it: 1️⃣ Testing code faster >> validating logic, challenging assumptions, and tightening SQL before I move ahead 2️⃣ Fixing broken code >> helping me debug issues faster and think through root causes, not just syntax 3️⃣ Designing better ETLs >> simplifying flows, reducing unnecessary complexity, and improving maintainability 4️⃣ Improving iteration speed >> when I am stuck, it helps me move from “what is wrong here?” to “here’s a better path” much faster ⚡ What I also like is that it supports the broader reality of data engineering work: ✅ SQL and Snowpark development ✅ Stored procedure refactoring ✅ ETL troubleshooting ✅ Streamlit app building ✅ Admin, governance, and cost-related workflows And that matters. Because real data engineering is not only about writing code. It is about building reliable pipelines, reducing rework, improving quality, and shipping faster without increasing risk. That is why COCO feels important to me. It is not just another AI assistant. It feels like a product that understands why enterprise data teams have been slow to fully embrace external coding copilots and offers a much more natural path forward within Snowflake. That is a big deal. Have you tried COCO yet? ---- ♻️ Repost if this resonates with your data journey 🔔 Follow Dewank to learn more about Data Science & Analytics!
-
Karthik Tadinada reacted on thisKarthik Tadinada reacted on thisFor 7 years, my business has grown entirely by word of mouth. I don't market, or advertise, or even have a website. That's partly because I get all the business I want without it, and partly because self-promotion still makes me really uncomfortable*. But after Satya's post yesterday, I figure I'm going to get 24 hours of uncomfortable levels of attention anyway, so I might as well embrace it... I'm marketing for the first time now because I've realized that what people come to me for - building complex models and tools - is not where I can actually be most useful. I started my career at the Boston Consulting Group (BCG) in London. It became apparent early on that I was good at Excel. Within 6 months, I was running the office-wide training. And pretty soon, I had a de facto drop-in clinic at my desk on Fridays. I'd help people get unstuck, or automate a manual process, or speed up a slow model. A few 10-minute interventions saved a massive amount of consultant hours. One strong user who can quickly help many others is a huge force multiplier. And that's what I'm offering now: custom Excel and modeling training targeted your company's force multipliers. Instead of hiring me to build one model, let me work with your top 1-3 power-users to raise their game, and you'll get the benefit of your whole company getting a little smarter with data. If you're interested in developing your team's internal experts, DM me. I'm looking to partner with a couple of firms to pilot this. And if you're wondering if AI has made this kind of training irrelevant - I would say firmly the opposite. AI can give a lot of leverage, but the quality of output is exponentially higher when the user knows what a world-class model looks like. So the force multiplier effect is only bigger. * Time spent in New York and on LinkedIn has clearly raised my tolerance, because I wouldn't have written this post a few years ago. I'm not sure that's a good thing, on balance...
Experience & Education
-
Fortify Solutions
View Karthik’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Languages
-
Hindi
Limited working proficiency
-
Telugu
Limited working proficiency
Other similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More