WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=oWqsBuhRRZ4 NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/oWqsBuhRRZ4.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:12.500 --> 00:00:14.500 We have a really exciting session for you guys. s3 00:00:14.500 --> 00:00:28.100 So, um you probably don't realize how many systems and how much of the infrastructure which drives modern finance, which drives all the things which happen in the financial district in New York City s4 00:00:28.100 --> 00:00:35.000 are driven off very old systems which can use a lot of modernization. s5 00:00:35.000 --> 00:00:48.300 And you probably interact with them on your on your cell phone, you maybe you're banking on a mobile app, but you don't realize that underneath the covers there's business rules, there's batch jobs running in nightly processes. s6 00:00:48.300 --> 00:00:52.400 You have all these different data contracts and things running under the covers. s7 00:00:52.400 --> 00:00:57.600 And it's running on infrastructure which people are frankly afraid to touch, right? s8 00:00:57.600 --> 00:01:03.700 If you if you make one little line of code change, you risk breaking the whole system. s9 00:01:03.700 --> 00:01:07.400 And um so this is um fun fact about myself. s10 00:01:07.400 --> 00:01:13.900 So, um you know, now I'm I'm known for doing a lot of developer relations, AI conferences, book writing. s11 00:01:13.900 --> 00:01:26.600 Um back 30 years ago, I was a Java developer who was migrating COBOL code off mainframes and like, you know, figuring out how to modernize applications. s12 00:01:26.600 --> 00:01:34.700 And it was multi-year projects, very challenging work which inevitably you hit little bugs and issues. s13 00:01:34.700 --> 00:01:43.300 You couldn't test it thoroughly enough to hit all the different corner cases and very very challenging projects which often failed. s14 00:01:43.300 --> 00:01:54.400 So, I think that what we're going to show you today is amazing cuz it's a new path forward where you can drive the change and you can actually do this, but s15 00:01:54.400 --> 00:01:58.300 um why don't you tell us a little bit about how this works, Neeraj? s16 00:01:58.300 --> 00:01:59.300 Thank you, Steve. s17 00:01:59.300 --> 00:02:03.292 So, we're going to do this with an example, right? s18 00:02:03.292 --> 00:02:10.492 Uh, in the context of a public demo system which Amazon developed a few years ago. s19 00:02:10.492 --> 00:02:12.892 This is the AWS card demo. s20 00:02:12.892 --> 00:02:22.592 And going to take a specific example here where we are trying to compute interest, right? s21 00:02:22.592 --> 00:02:37.292 And so, as you can see in this little chart, we have a batch job that's calling the interest program which is getting the copybooks and layouts and then generating basically VSAM s22 00:02:37.292 --> 00:02:38.592 data, right? s23 00:02:38.592 --> 00:02:53.892 And so, the output here is money and of course, the result depends on the inputs, the job that runs it, the account updates, not just the formula for computing the interest, right? s24 00:02:53.892 --> 00:02:56.592 So, let's remember this example. s25 00:02:56.592 --> 00:03:04.392 We're going to keep coming back to this, hit that theme over and over again as we progress through the talk. s26 00:03:04.792 --> 00:03:17.292 So, when I want to modernize this and I tell AI, hey, change that formula from Cobalt to Java, all it's going to do is find the formula, try to convert that, right? s27 00:03:17.292 --> 00:03:25.192 What it doesn't find is the the record precision, the account updates, the transaction output, the batch contract. s28 00:03:25.192 --> 00:03:31.392 And if you miss any of these, you're not guaranteed to get the right outcomes. s29 00:03:32.792 --> 00:03:37.692 So, the failure here is not that AI can't write the code. s30 00:03:37.692 --> 00:03:45.292 It's that AI doesn't know the 14 other places where that logic, that code is being used. s31 00:03:45.292 --> 00:03:46.792 Okay, and this this looks pretty scary. s32 00:03:46.792 --> 00:03:51.492 I mean, besides the AI not knowing the code, that pipe looks like it's about to burst. s33 00:03:51.492 --> 00:03:54.992 Yeah, and of course, the AI doesn't know that it's going to happen any moment now. s34 00:03:54.992 --> 00:04:01.292 So, Stephen, you would think that a context expansion would fix this. s35 00:04:01.292 --> 00:04:02.894 Why doesn't that s36 00:04:02.894 --> 00:04:03.094 Yeah. s37 00:04:03.094 --> 00:04:15.294 So I I think that one of the things actually the previous presenter talked a bit about this, how we're we've all been relying on better models, on having better systems, a better AI systems to solve this. s38 00:04:15.294 --> 00:04:20.694 And frankly, it hasn't been working for these sorts of use cases and the reason is context. s39 00:04:20.694 --> 00:04:28.294 So you can do vector databases to get and you know expand your context, but all it gives you is similarity searches. s40 00:04:28.294 --> 00:04:34.394 You can do relational databases as well and this gives you a much bigger repository. s41 00:04:34.394 --> 00:04:43.094 You can hook it up via an MCP server, but again, you're you're limited on kind of single hop and like looking up facts and information by SQL queries. s42 00:04:43.094 --> 00:04:51.794 What a graph gives you is a graph gives you this multi hop accurate information and context for your model. s43 00:04:51.794 --> 00:05:03.894 And the reason why this context is so important is because your model cannot possibly give you the right results because it can't utilize its own context. s44 00:05:03.894 --> 00:05:09.694 So most models, you know, nowadays are advertised with a million token context windows. s45 00:05:09.694 --> 00:05:15.094 Um but what do they actually how much usable context do you actually get? s46 00:05:15.094 --> 00:05:18.594 And the reality is you get 200k or less usable context, right? s47 00:05:18.594 --> 00:05:30.194 So it's by by study the accuracy goes down dramatically as you fill the context more and this means you have to give precise and accurate context. s48 00:05:30.194 --> 00:05:39.794 Yeah, the effective context these days is about like 20 to 30% which means about 200 to 300,000 lines of code. s49 00:05:39.794 --> 00:05:42.194 If your software is bigger than that, too bad. s50 00:05:42.194 --> 00:05:44.994 You're going to start losing information. s51 00:05:45.894 --> 00:05:50.594 So context is serial, but information has relationships. s52 00:05:50.594 --> 00:05:53.594 Information has connections between it. s53 00:05:53.594 --> 00:05:59.794 And this is the key to how you can actually line up your data. s54 00:06:03.594 --> 00:06:04.294 Yeah. s55 00:06:04.294 --> 00:06:06.995 So I mean here so code becomes a graph, right? s56 00:06:06.995 --> 00:06:10.695 You can see all of these relationships that are being mapped here now. s57 00:06:10.695 --> 00:06:15.395 Uh, and like the the program is there. s58 00:06:15.395 --> 00:06:21.895 It has the the job that's executing it, the copy book that are getting included, so on and so forth, the outputs. s59 00:06:21.895 --> 00:06:27.295 The important part here is the stuff on the right. s60 00:06:27.295 --> 00:06:32.295 So, every edge in the graph keeps its identity, right? s61 00:06:32.295 --> 00:06:33.895 Where did it come from in the repo? s62 00:06:33.895 --> 00:06:34.995 What revision was it? s63 00:06:34.995 --> 00:06:36.595 What symbols does it touch? s64 00:06:36.595 --> 00:06:41.595 There is evidence for the file and the and the source span, how it was extracted. s65 00:06:41.595 --> 00:06:48.295 And the graph keeps getting updated on on every merge as the code changes, so does the uh the graph. s66 00:06:48.295 --> 00:06:54.395 And that way you keep it fresh, you keep it current, you invalidate stuff that is no longer valid. s67 00:06:54.395 --> 00:06:57.595 Yeah, and you can see that the code is very much a graph-shaped problem. s68 00:06:57.595 --> 00:07:03.295 So just from the structure and the relationships, it's something where you have a lot of interconnects. s69 00:07:03.495 --> 00:07:08.995 Um, and then when you what why don't you talk about how this works for agents? s70 00:07:08.995 --> 00:07:09.595 Sure. s71 00:07:09.595 --> 00:07:16.495 So, each agent uh in Blitzz has a specific task context packet, right? s72 00:07:16.495 --> 00:07:23.595 And that includes what the task description is, uh but also where does it belong to in that graph, right? s73 00:07:23.595 --> 00:07:25.695 There's the slice of the source, right? s74 00:07:25.695 --> 00:07:36.095 What constraints it has imposed, what evidence it has available, and what questions are still not answered that that agent is now supposed to work on. s75 00:07:36.095 --> 00:07:38.495 Uh, and this is what we mean by infinite code context, right? s76 00:07:38.495 --> 00:07:44.595 It's not remembering the entire uh information in in your code base, in your software estate. s77 00:07:44.595 --> 00:07:52.295 It's basically having access to the right information at the right time, in the right place. s78 00:07:53.795 --> 00:08:00.095 So, Blitzz effectively works like a compiler for the entire knowledge in the enterprise, right? s79 00:08:00.095 --> 00:08:04.595 You start off by reverse engineering the entire software estate. s80 00:08:04.595 --> 00:08:09.497 You extract the business logic, uh the workflows, the dependencies. s81 00:08:09.497 --> 00:08:14.897 Then you get the requirements and the rules from the user to say, what do you want to do with this? s82 00:08:14.897 --> 00:08:19.697 And you convert that into an actionable agent action plan, right? s83 00:08:19.697 --> 00:08:25.097 This is the spec that you're going to use to make that forward change. s84 00:08:25.097 --> 00:08:28.197 Note that next block there, there is a human approval, right? s85 00:08:28.197 --> 00:08:35.897 Nothing happens without a human in the lead making decisions at every key point, right? s86 00:08:35.897 --> 00:08:44.397 But once you have that plan reviewed and approved by the human, now the agents go off, they do the code generation. s87 00:08:44.397 --> 00:08:47.997 But there is also validation that is baked into it at every step. s88 00:08:47.997 --> 00:08:58.297 And so the end result is you get tested and validated code in the form of a pull request and an evidence documentation for what has happened and how it has been validated s89 00:08:58.297 --> 00:09:00.597 in the form of a project guide. s90 00:09:00.597 --> 00:09:07.697 So there is a human approval at every step that matters. s91 00:09:07.797 --> 00:09:19.297 And because you have access to the full repository and you also have the instructions about how to build the environment, how to run the code, you are not restricted s92 00:09:19.297 --> 00:09:21.897 to just static information. s93 00:09:21.897 --> 00:09:27.097 You can actually observe the runtime behavior and make changes accordingly. s94 00:09:28.197 --> 00:09:32.097 So the the knowledge graph basically drives all of this, right? s95 00:09:32.097 --> 00:09:37.197 But that's not the only thing that helps here. s96 00:09:37.197 --> 00:09:41.697 There are two other things that compound the impact that this has. s97 00:09:41.697 --> 00:09:43.397 One is hyperscale orchestration. s98 00:09:43.397 --> 00:09:47.397 Like we have thousands of agents who all share that plan. s99 00:09:47.397 --> 00:09:54.197 Every agent has access to the effective context and there is nothing lost between them. s100 00:09:54.197 --> 00:09:57.797 And couple that with extreme validation, right? s101 00:09:57.797 --> 00:10:02.197 You have testing happening at every stage of code generation. s102 00:10:02.197 --> 00:10:06.897 We actually spend more computes on testing than we do on code gen. s103 00:10:06.897 --> 00:10:21.198 And this results in what is effectively production safe code, that has been tested, validated at enterprise scale and so it is safe to make merges into your production systems. s104 00:10:22.698 --> 00:10:25.698 So the agent action plan is is the spec, right? s105 00:10:25.698 --> 00:10:37.498 That's the plan before you start doing any actual work and it decides what the scope is, what gets transformed and what must be preserved, right? s106 00:10:37.498 --> 00:10:49.598 And a domain expert, a human would review this, make sure that it's aligned with their expectations, approve it and this is coupled with the four controls that are inherent s107 00:10:49.598 --> 00:10:50.898 at every step. s108 00:10:50.898 --> 00:10:55.498 There is system level context, that's the knowledge graph we talked about, right? s109 00:10:55.498 --> 00:10:59.498 Every change impacts the entire estate, so everybody needs to be aware of that, right? s110 00:10:59.498 --> 00:11:03.698 There is continuous validation, so testing, testing, testing, right? s111 00:11:03.698 --> 00:11:05.398 There's no substitute for that. s112 00:11:05.398 --> 00:11:11.298 You have explicit guardrails in place to address your security, your governance, your compliance requirements. s113 00:11:11.298 --> 00:11:15.898 These are inherent to any regulated industry and of course finance, right? s114 00:11:15.898 --> 00:11:20.698 And then the human in the lead because the decisions that matter are owned by real people. s115 00:11:20.698 --> 00:11:23.998 The AI is not making them autonomously. s116 00:11:25.098 --> 00:11:28.798 And this of course preserves the behaviors, right? s117 00:11:28.798 --> 00:11:39.698 The existing behaviors have to be transferred as is to the to the modernized code and how do you verify that? s118 00:11:39.698 --> 00:11:40.798 Right. s119 00:11:40.798 --> 00:11:47.298 So in this case like we want to preserve the money like the precision, the rounding. s120 00:11:47.298 --> 00:11:54.398 We want to preserve the state, the balances and the transaction records, failures, operations, everything, right? s121 00:11:54.398 --> 00:11:59.298 If we have defects, they also need to be managed with explicit decisions. s122 00:11:59.298 --> 00:12:04.898 The rules have to be preserved as well and you have to prove the parity. s123 00:12:04.898 --> 00:12:09.698 You can't take it as an assumption or as a promise. s124 00:12:09.698 --> 00:12:14.399 So the way we do that is there is testing on the baseline. s125 00:12:14.399 --> 00:12:26.299 There is equivalent testing happening on the modern system, and we are looking at inputs and outputs, the returned values, what states get persisted, what is the transaction output look like, right? s126 00:12:26.299 --> 00:12:28.199 What are the error outcomes? s127 00:12:28.199 --> 00:12:33.299 Not just the happy path, but how do you deal with like bad situations? s128 00:12:33.299 --> 00:12:43.399 And all of those have to be exactly matching in the context of each operation, and the differences have to be none. s129 00:12:43.399 --> 00:12:47.199 That's our promise to deliver this. s130 00:12:47.899 --> 00:12:49.599 So this is layer, right? s131 00:12:49.599 --> 00:12:51.299 This doesn't happen only in the end. s132 00:12:51.299 --> 00:12:53.899 This is baked into the very fabric of the code generation. s133 00:12:53.899 --> 00:12:58.799 Every line of code that gets generated gets built and tested with unit tests. s134 00:12:58.799 --> 00:13:01.099 The build has to be compiled. s135 00:13:01.099 --> 00:13:07.999 Runtime integration, it has to fit within the existing kind of framework that you have. s136 00:13:07.999 --> 00:13:14.999 Your data parity has to be there on every write, and the end-to-end workflow has to be identical. s137 00:13:15.299 --> 00:13:18.799 So this is a code from one of the customers where we actually did this, right? s138 00:13:18.799 --> 00:13:23.799 The transpiled job produces output that is byte for byte identical to the golden mainframe baseline. s139 00:13:23.799 --> 00:13:27.299 This is from a Fortune 100 insurance. s140 00:13:27.399 --> 00:13:29.199 Here's a fun example. s141 00:13:29.199 --> 00:13:37.899 So so this is something I can get into as an old Java developer where we we made lots of errors with big decimals and rounding in our implementation. s142 00:13:37.899 --> 00:13:40.399 So AI can do this better, right? s143 00:13:40.399 --> 00:13:40.499 Right. s144 00:13:40.499 --> 00:13:43.099 I mean you you you would think so, right? s145 00:13:43.099 --> 00:13:50.199 Because like you computed a a Java version of this this formula to compute interest, right? s146 00:13:50.299 --> 00:13:51.799 Cobol behaves differently than Java. s147 00:13:51.799 --> 00:13:55.799 Java has an inherent rounding up built into it. s148 00:13:55.799 --> 00:14:06.599 What that means is the modern code compiles, it generates results that you believe are correct because it's behaving like normal Java code, right? s149 00:14:06.599 --> 00:14:12.299 But you have this little difference of that 1 cent that happened because of the inherent rounding. s150 00:14:12.299 --> 00:14:15.999 If you didn't have that parity test, you wouldn't have caught this. s151 00:14:15.999 --> 00:14:19.401 And now imagine this happening over a million transactions. s152 00:14:19.401 --> 00:14:21.401 Fun. s153 00:14:22.701 --> 00:14:27.401 So that was just one kind of simple example, right? s154 00:14:27.401 --> 00:14:33.101 But now you can think of like a full trade posting engine where this gets applied throughout the entire workflow. s155 00:14:33.101 --> 00:14:44.501 This is something that we did for a tier one bank where we had a monolith that had stored procedures uh cross database uh kind of uh write access going on. s156 00:14:44.501 --> 00:14:55.801 This got converted into six capability modules uh with PostgreSQL, Java, TypeScript, Python, the fully modern stack uh and with full parity. s157 00:14:56.001 --> 00:14:56.601 Right? s158 00:14:56.601 --> 00:15:05.401 And took only 18 days uh whereas their original plan was to have uh 18 engineer weeks dedicated to this project. s159 00:15:05.401 --> 00:15:05.501 Right? s160 00:15:05.501 --> 00:15:09.801 This was doing things the kind of human way. s161 00:15:09.801 --> 00:15:12.401 Uh with Blitzz, it got reduced to 18 days. s162 00:15:12.401 --> 00:15:12.601 Right? s163 00:15:12.601 --> 00:15:14.401 We We wrote 4,000 plus tests. s164 00:15:14.401 --> 00:15:24.301 Uh the modern code was less than half the size of the of the new of the old stack uh and the customer was happy. s165 00:15:24.301 --> 00:15:30.001 There is no doubt that we can use Blitzz to modernize the whole system is is what they said. s166 00:15:30.601 --> 00:15:34.001 Another example, mapping a 35 million line mainframe. s167 00:15:34.001 --> 00:15:41.301 Uh this is something uh that they had spent on an SI about eight months, several million dollars and had nothing to show for it. s168 00:15:41.301 --> 00:15:46.701 Uh Blitzz reverse engineered that in a matter of four days and this is what they have to say. s169 00:15:46.701 --> 00:15:50.901 They've been in the dark room with a pencil light inspecting each individual module. s170 00:15:50.901 --> 00:15:55.501 Blitzz feels like flipping on the overhead light, seeing the whole room at once. s171 00:15:55.501 --> 00:16:01.401 This is what AI accompanied by a knowledge layer can do for you. s172 00:16:01.501 --> 00:16:02.401 Right? s173 00:16:02.401 --> 00:16:06.201 Uh another example, we modernized a 25 million line monolith. s174 00:16:06.201 --> 00:16:17.801 This was a policy admin platform again uh compared to a human aided by AI like your typical IDE based uh agentic tools uh would have taken a certain amount of time. s175 00:16:17.801 --> 00:16:19.502 We did it nine times faster. s176 00:16:19.502 --> 00:16:27.002 85% of the code was autonomously completed and almost 90% of that made it into production. s177 00:16:27.402 --> 00:16:30.402 This extends to different systems, different companies. s178 00:16:30.402 --> 00:16:36.902 As long as you understand the whole system before you change it using AI, you're good. s179 00:16:37.702 --> 00:16:39.902 And this doesn't just stop at modernization, right? s180 00:16:39.902 --> 00:16:42.202 Ultimately, transformation is the goal. s181 00:16:42.202 --> 00:16:44.502 Britezi is not a modernization platform. s182 00:16:44.502 --> 00:16:46.402 It is a transformation platform. s183 00:16:46.402 --> 00:16:49.302 You can do new product development. s184 00:16:49.302 --> 00:16:50.402 You can add new features. s185 00:16:50.402 --> 00:16:53.202 You can do security remediation and detection. s186 00:16:53.202 --> 00:16:58.502 All kinds of use cases open up when you have that entire estate available to you. s187 00:16:59.102 --> 00:17:09.002 And so another coming back to the card demo, now that you have modernized it, you may want to add a new feature, a promotional rate to the interest path, right? s188 00:17:09.002 --> 00:17:14.602 You already have the upgraded knowledge path now because you already have the modern system. s189 00:17:14.602 --> 00:17:16.802 Your knowledge layer got updated. s190 00:17:16.802 --> 00:17:18.702 You have your governance rules established. s191 00:17:18.702 --> 00:17:20.502 You have your test suite expanded. s192 00:17:20.502 --> 00:17:27.602 So now this becomes a lot easier and faster and cheaper to do than it had been before. s193 00:17:28.902 --> 00:17:44.202 And then you can do this beyond the code to expand this kind of knowledge to apply to new cases, new use cases like providing audience, providing evidence for like audit ready s194 00:17:44.302 --> 00:17:46.802 compliance issues. s195 00:17:49.102 --> 00:17:56.202 So as you can see, this is heading beyond software and mapping the whole business now. s196 00:17:57.102 --> 00:18:05.102 Yeah, so I think that you can see the power of building a an modern enterprise layer on top of a knowledge graph. s197 00:18:05.102 --> 00:18:18.902 And it starts with a graph, builds out the business rules, business processes, and then allows you to build end-to-end systems which not only solve the modernization use case, but they also give you capabilities s198 00:18:18.902 --> 00:18:20.302 to expand your business. s199 00:18:20.302 --> 00:18:29.303 So you can now do um for example, Dora compliance on your um entire system to make sure it meets regulatory requirements. s200 00:18:29.303 --> 00:18:31.303 You can track business rule changes. s201 00:18:31.303 --> 00:18:47.303 You can do financial fraud and build um use cases which will track down other sorts of problems and this expands to even further out where you can extend your system and build more value off of it. s202 00:18:47.303 --> 00:18:51.703 We want to leave you with a few references to um to take with you. s203 00:18:51.703 --> 00:19:05.103 So um one is the Nodes conference, that's the QR code on the left, is a full 24 hours of sessions learning on knowledge graphs on how you can build and apply a bunch of these concepts s204 00:19:05.103 --> 00:19:08.103 for our knowledge graphs to your own systems. s205 00:19:08.103 --> 00:19:10.003 And um Niraj, you want to talk about s206 00:19:10.003 --> 00:19:17.503 Yeah, the Blu Age sandbox, that's the QR code on the right, will allow you to in fact bring in your own code and uh see this for yourself. s207 00:19:17.503 --> 00:19:19.103 Don't just take our word for it. s208 00:19:19.103 --> 00:19:26.603 You can also see the AWS card demo modernization that we showed here uh as an example there. s209 00:19:26.603 --> 00:19:30.403 Okay, so thanks so much for joining our session at A and G in New York City. s210 00:19:30.403 --> 00:19:35.903 And um if you have any more questions, feel free to drop by the Neo for J booth or come up and chat with us after the talk. s211 00:19:35.903 --> 00:19:36.503 Thank you. s212 00:19:36.503 --> 00:19:38.903 Thank you very much.