Feb 27, 2026 Divami / Vijay - Transcript 00:00:00 Vijay Kamineni: Okay. Yeah. Naveen Puttagunta: When I talk about lineage, I'm I'm kind of imagining that there are, you know, four stages of processing and at each stage, you know, it is being tagged with okay, it came from this machine, it came from this method, uh, you know, the source of the material is uh this uh vendor, etc., etc., etc., right? So, it is tagged with the lineage of the production line. So then you know then the uh fault analysis will be much much richer because now you can uh basically whatever questions you're asking here you can ask about the faults you know uh you can identify insights about uh what particular lineage is causing more problems etc. So if if this thought process is uh reasonable then I wanted Vijay Kamineni: Uh-huh. Naveen Puttagunta: to kind of get insight into okay typically in the anominium you know either the uh either the heat plant or whichever example that you take typically what does the production uh staging look like right so this by the way this is you know we I got this more from uh we did the analysis as I was telling you we did this analysis for blue yonder supply chain lineage right we are able to you know look through the data and understand the supply chain within a day right like you know we understand the complete supply chain we can give 00:01:43 Naveen Puttagunta: you the supply chain graph u you know that came out really really well actually uh when we worked with that blue yonder supply chain example so similar to supply chain we can look at the production line and and kind of show you the lineage maybe uh then we can start you I know working with that and and looking at visualizations. But if that thought process is right, maybe you can uh help say oh typically in a you know aluminum production line it is not unreasonable to think there are three stages. uh in each stage there is a machine and there is a a person or uh uh you know team something like that where if that is the case then we will go um what do you call uh tag that uh uh that data that fault data that we have with that uh lineage of the uh production line then I know so that that tagging is going to be synthetic data it's just a simulation that's okay so then you can start asking these questions about the production line and analysis and all of that. 00:02:50 Naveen Puttagunta: Does that sound reasonable or is it like completely made up? Vijay Kamineni: No. uh it's reasonable for us to test our capability internally but for presentation it'll be oneonone that'll be very specific for one user base one com one or two companies right if you get that specific so we want to take the concept of a machine sequence is a machine sequence No two plants will have the same machine sequence. Naveen Puttagunta: right? Vijay Kamineni: So um even within Logan aluminum we have uh four what we call direct caster casting furnaces but two of them are alike. Naveen Puttagunta: What Vijay Kamineni: Three and four are completely different from all others. Naveen Puttagunta: happened? Vijay Kamineni: So they're not even within a one plant the so it's not as standard as you think in tech technology and uh retail and those are standardized manufacturing that's the problem and the opportunity that you can't standardize it like that when once you get that once you get to I Naveen Puttagunta: It's okay. Vijay Kamineni: like what you're thinking that's what the client wants but everybody wants it for their machine sequence. 00:04:27 Vijay Kamineni: So, Naveen Puttagunta: Yes. Vijay Kamineni: we can use one of those and show what you're asking to show. Um, Naveen Puttagunta: So is your concern that if we show if we pick one such thing they will think that oh this will only work for one such thing. Vijay Kamineni: Yes. Yes. Naveen Puttagunta: I Vijay Kamineni: Manufacturing manufacturing mindset is uh if it's not built inside here, Naveen Puttagunta: see Vijay Kamineni: it doesn't work. That's why a lot of people didn't go for productizations because everybody thinks they're extremely unique and u Microsoft Dynamics and SAP won't work for us because we are so unique. They're working from government to warehouses to hospitals in large organization but manufacturing things it won't work. Um so that's a mindset we want to work with. Naveen Puttagunta: Hey, Vijay Kamineni: That's why we try to avoid that level of specificity especially how when we're entering a very wide market as we create niches then we speak specifics. Naveen Puttagunta: what's happening? Vijay Kamineni: We haven't created those niches niches yet. We don't want to because if we do we don't know where the first problem who uses first to start working on that problem. 00:05:38 Naveen Puttagunta: Okay. Vijay Kamineni: Once we once we get that we build from we build adjacent and vertical on that problem. Whoever gives you the first contract will be what we call your beach head customer right and we'll start working on that and who like trace diecast if they give a Naveen Puttagunta: Heat. Heat. Vijay Kamineni: problem there are two other die casting companies in Kentucky so they'll have the same problem so we go and expand that way and then we'll go to trace diecast suppliers supply chain Naveen Puttagunta: Heat. Vijay Kamineni: and expand vertically and then if we show trace diecast example the immediate upstream and downstream supply chain can understand it because oh if it works for them it probably will work for us. So even consulting companies struggle with uh I think they brema and they Naveen Puttagunta: Okay. Vijay Kamineni: they'll meet uh what we call manufacturing extension partnership. It's a it's a uh national uh small Naveen Puttagunta: Okay. Vijay Kamineni: manu small and mean medium manufacturer problem solving. It's a NIST uh initiative. They so we'll be meeting them in Kentucky. 00:06:53 Vijay Kamineni: It's a Kentucky chapter. There are every state has a chapter. Every state should have one and only one of those chapters helping manufacturing. So even they have the same problem. Naveen Puttagunta: You Vijay Kamineni: So uh we'll I don't mind doing it but we might be Naveen Puttagunta: okay? Vijay Kamineni: c putting ourselves into a corner. Uh what we have here these are the standard questions right? It doesn't matter we should get into the standard questions and say everybody has a part everybody has these metrics. Sorry if I feel like I'm pushing back. Naveen Puttagunta: Okay. Vijay Kamineni: I don't want to Naveen Puttagunta: No, no, that that's fine. Vijay Kamineni: uh Naveen Puttagunta: I mean, that's valuable feedback, right? If if it's if it's actually going to backfire showing the detail and all of that, then might as well not do that. I mean, I I understand because you are a better judge of how they think and what the mindset is. uh you know in the tech world right we're used to showing a demo and saying hey your problem is just like this but you know it's slightly different they're able to do it and they will accept it because they understand that you know if you're able to do it for one then you're able to do it for other but I understand that you know uh these guys are mindset is not like that right they we but you're right we absolutely want to show the 00:08:09 Vijay Kamineni: Yeah. Naveen Puttagunta: capability of being able to do solve for anything uh not that oh we're trying to sell a product just for on one thing. Vijay Kamineni: Right? Naveen Puttagunta: So Vijay Kamineni: If we we're doing product, we have to keep it a we have to keep it with ourselves. Naveen Puttagunta: understood. Vijay Kamineni: We might be cop we might be doing the same thing for three of them but they should feel we did it only for them for the for the business to come Naveen Puttagunta: Yeah. Yeah. Yeah. Vijay Kamineni: in. Naveen Puttagunta: Got it. Vijay Kamineni: So, Naveen Puttagunta: So So on in that case, Vijay Kamineni: um Naveen Puttagunta: right, let me ask you a a different question. then what can I embellish this uh fault data with so that uh somebody can ask so I mean see the whatever you're playing with right now that's that's one of them that's already there that's fine now I have the fault data for let's say 5,000 heat images right there are faults in 5,000 of those images there are coordinates there are image data I have all of that do you think I can provide something uh meaningful um uh uh you know similar things like you know okay you you are able to ask and uh it's able to answer so what can I uh um 00:09:32 Vijay Kamineni: So at at this point at this point if I have that I would ask questions Naveen Puttagunta: yeah Vijay Kamineni: about the defect classifications rather than tying out to the production because we don't have the tied data we can always start talking about but in your Naveen Puttagunta: Heat. Vijay Kamineni: process uh we can tie them but we can tie it to the machine if you have the tags we can tie it to. So we talk about the possibility and we ask with uh we can stay with defect Naveen Puttagunta: Heat. Vijay Kamineni: classification. How many types of defects are there? Right? It it learned something. It it learned it learned by it's a it's a self-arning model. So what is the C means classification on that? Right? We won't ask it like that. We'll ask it in English. How many defect groups and which ones are which ones are more uh are repetitive versus one or one off. Naveen Puttagunta: Bye. Vijay Kamineni: So in the repetitive ones, how many of them? 00:10:29 Vijay Kamineni: Then we'll we're looking at trying to identify which defect we want to target to solve to reduce. Naveen Puttagunta: Amazing. Vijay Kamineni: So then we can go into probably give some easier things that we can add is cost per defect tag and we can make it up uh Naveen Puttagunta: So Vijay Kamineni: and say okay you have a scratch and you have a smudge. Naveen Puttagunta: good. Vijay Kamineni: I'm making up making those. So if scr if you have 300 30 scratches and four smudges but each smudge is $300 and a scratch is only a dollar. Naveen Puttagunta: Heat. Vijay Kamineni: Solving even one smudge is better than solving a whole bunch of Naveen Puttagunta: Heat. Vijay Kamineni: scratches. That's how we'll think on the data you have. So we are tying it to a result but we're not tying it to a machine processor. It's not that kind of tying. So we can come up with some here is what I'm thinking given the time the closeness we have with the time uh when Sam and I are Pratima will talk to somebody on Monday my here is what I'm thinking tell me if it doesn't work for some reason my we'll ask some data from 00:11:36 Naveen Puttagunta: Okay. Vijay Kamineni: that when we they'll collect some data but we'll ask some data that is that we can feed enterprise brain So after dinner I'm scheduling my time after dinner for three Naveen Puttagunta: Okay. Vijay Kamineni: hours we'll have the session like we had yesterday and your team will understand they they'll they just came to the by the time Fatima and the wakes up they should be able to run we Naveen Puttagunta: Okay. Vijay Kamineni: know the specific question they're trying to answer and the data they have we can do a we can run it through enterprise brain and say okay with the data you have when we ask the question it got us some answers It didn't give all the answers. So we might have to bring in these data and here is how it will work. Here is how far. So do you want to ask any question that will give them a better experience than trying to do what we are trying we'll get the same result. They'll have something that they can experience but it's coming from one question one data set. 00:12:43 Vijay Kamineni: I think we can do it in a I saw you do it in a day. So in a day or two we can turn it around and if if not most of them will still be in Naveen Puttagunta: Yeah. Vijay Kamineni: the come will come for the workshop. So after the workshop we can show oh here are the three data we collected within the week and here is how fast we came to that. That's a better capability. So your team if it's standing by probably that's one way I'm thinking that's Naveen Puttagunta: Heat. Vijay Kamineni: one way I'm thinking how to make them feel we are we have Naveen Puttagunta: Okay. Vijay Kamineni: better capability than what they otherwise would find correct me might not be for some Naveen Puttagunta: Fair enough. Yeah. Yeah. No, no. Uh no no that that's fine. Actually if there is a data right because we're asking for data in data dumps in excels or CSVs or whatever generally it will not be you know hundreds of tables right it'll 00:13:40 Vijay Kamineni: Yeah. Naveen Puttagunta: typically be maximum 10 table 10 20 tables so 10 20 tables we Vijay Kamineni: Yeah. Yeah. Naveen Puttagunta: can you know process it through a day so you do it you know by if we get the data in the evening by your morning we can have uh enterprise brain kind of trained on that data so Vijay Kamineni: Yeah. Naveen Puttagunta: it Vijay Kamineni: Or if it takes good because they're not going back to the same people in the next day. But even if it takes a little longer, I think we can do that. Let's see if we can even do one of Naveen Puttagunta: Okay. Vijay Kamineni: those. Naveen Puttagunta: Okay. Sounds good. So I mean you've you've met Pratima. She's not going to let it go if I say, "Oh, we're not going to do visual data." So I'm going to do Yeah. Vijay Kamineni: No, we'll do it. Let's Let's Let's load it up. Naveen Puttagunta: So what I'm going to do is Yeah. Vijay Kamineni: Yeah. Naveen Puttagunta: Go ahead. Go ahead. 00:14:25 Vijay Kamineni: Um do it uh add a cost column, Naveen Puttagunta: Okay. Uh Vijay Kamineni: add a cost column uh and um if you can add a Naveen Puttagunta: okay. Vijay Kamineni: date when the when the defect occurred and then put some Naveen Puttagunta: Okay. Vijay Kamineni: cluster some defects in one period. You're putting fake dates anyway. You cluster them in uh if you if you spread it over last 15 days. Naveen Puttagunta: Yeah. Vijay Kamineni: say last Saturday was the defect and or every Saturday you have more defect. You make up you make up those stories and then let it show the data. So add a date to it. Add a uh if you want to add uh location or machine usually one type of defect happens only on one machine. Naveen Puttagunta: right? I see. Vijay Kamineni: So if if they're if they're taking a picture of the backside of the sheet, so that just went through one machine. So that all the 300 images will be from that step like for like because when it goes to the next step it'll change form and it's a different picture. 00:15:31 Naveen Puttagunta: Nice Vijay Kamineni: So so assume okay all of these defects came from this station and Naveen Puttagunta: thing. Vijay Kamineni: uh you can say you can go ahead and put names operator one operator two operator three operator number 1 2 3 4 next. Naveen Puttagunta: Heat. Heat. Vijay Kamineni: So create four operator numbers or team numbers and say now you have a cost per of the defect. You have a time when it occurred. You have the operator when it occurred and then cluster them and say this one team is having more defects of smudges than the other team. So we can make that up and that's more interesting for them than what I mean that's one way to get Pratima a good story to tell and say here the your uh um visual um I call it Naveen Puttagunta: Got it. Vijay Kamineni: computer vision I'm trying to learn your your enterprise uh so Naveen Puttagunta: No, no, no. Computer vision is fine, too. Yeah. Vijay Kamineni: our computer vision model has identified these now we're asking questions and it has this data of when it happen and how we can come to okay this is what we should solve this team has to be trained and or Saturday something is happening on Saturdays I'm making it this up then we'll solve for Saturday we won't have the team on Saturday or Saturday we might need supervision or a team needs training or um 00:16:52 Naveen Puttagunta: or something is going on on Saturday, right? The machine is taking a holiday. Vijay Kamineni: right yeah or uh we have uh time of the day is a great Naveen Puttagunta: Yeah. Vijay Kamineni: this thing we find uh issues happen uh at night shift more issues in night shift than dayshift. So those are the kind of things you can make up and those actually happen all Naveen Puttagunta: Okay. Vijay Kamineni: the Naveen Puttagunta: Well, Satyasri Prabhakar Mantripragada: Vually will there be any acceptance or rejection criteria defined Naveen Puttagunta: I see. Vijay Kamineni: time. Satyasri Prabhakar Mantripragada: for the uh let's say the coil or the sheet based on the number of bugs Naveen Puttagunta: Heat. Satyasri Prabhakar Mantripragada: or the type of bugs or the severity of the bugs. As you said smud is more important over a scratch. Vijay Kamineni: Yeah. Yeah. You you if you show me the images, I can I can make it up, but I'm making it up. I'm not telling the because those images are not actually what's happening in any plant, Naveen Puttagunta: Heat. 00:17:46 Vijay Kamineni: any one plant. So, uh yeah, Naveen Puttagunta: You're Vijay Kamineni: you you assign cost or cost or value for a defect. These defects you can't ship to the customer. Naveen Puttagunta: crazy. Vijay Kamineni: of these defects you can rework reworkable or non-reworkable defects right good good thinking that's the kind of thinking you should ask so those are the columns you should add and say now we are slicing and dicing the results after it classified we're trying to get Naveen Puttagunta: Heat. Satyasri Prabhakar Mantripragada: Amen. Vijay Kamineni: business answers from that um it's a it's an interesting problem you all have picked up on and Naveen Puttagunta: Heat. Vijay Kamineni: uh I I'll give you I'll give you a real life example that somebody is trying to do uh sidewalk inspection. Satyasri Prabhakar Mantripragada: Thank you. Naveen Puttagunta: Let's Vijay Kamineni: They haven't solved it. Cities cities uh spend a whole b I mean millions of dollars on uh insurance claims when people fall on sidewalks. So there is one company which created a which created a sidewalk. Yeah. There's a camera that goes on the sidewalk. 00:18:55 Naveen Puttagunta: go. Vijay Kamineni: Somebody will drive drive it through the sidewalks and capture a video and it should identify different types of cracks in the sidewalk and whether the crack is trip hazard or just a crack that needs to be or just a a Naveen Puttagunta: What Vijay Kamineni: beauty beauty problem. So that so yes vision is always already solved but Naveen Puttagunta: patient? Vijay Kamineni: nobody is solving for that and if you go on Google there might have prototypes nothing is being used in any city Naveen Puttagunta: Okay. Vijay Kamineni: and it should not be a it's it's just a mindset problem. Pizzas are being delivered by a automated bot. Why can't you get a image from an automated bot and classify it on a trip hazard? How can why are you not identifying a trip hazard for cities? It's a mindset difference. But um so I I'll play with this. Naveen Puttagunta: Let's Vijay Kamineni: You all play you ask me questions on we got these three columns. Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: We want two more columns. VJ, can you look at it and give another dimension to that? 00:20:12 Vijay Kamineni: We'll find dimensions. I can make up dimensions. Chpd can make up good dimensions to it costsaving Satyasri Prabhakar Mantripragada: Yeah. Naveen Puttagunta: see. Satyasri Prabhakar Mantripragada: Some of the things that we were looking at is um what is your first time yield uh Vijay Kamineni: dimensions Satyasri Prabhakar Mantripragada: versus uh uh the target um actually the target versus what is the Vijay Kamineni: on the vision system. Satyasri Prabhakar Mantripragada: current production rate. Uh those are Naveen Puttagunta: Thank Vijay Kamineni: On the vision system P. Satyasri Prabhakar Mantripragada: the Vijay Kamineni: No, not on the vision system P. Naveen Puttagunta: you. Vijay Kamineni: Right. Satyasri Prabhakar Mantripragada: uh we thought we'll be able to make it but uh we should be able to define when is the fastest first time Vijay Kamineni: Yeah. Satyasri Prabhakar Mantripragada: yield. Vijay Kamineni: When is the first time this defect material uh appeared? Right. If you put the date on if you to put the date column, that's a good when when did it start? It was these defects started off in the last three days. 00:21:04 Vijay Kamineni: Yeah, that's a good that's a good question to ask. If you create the data on the on the date set like Satyasri Prabhakar Mantripragada: And in case of a scenario where live monitoring happens via on over the conveyor Naveen Puttagunta: I think Satyasri Prabhakar Mantripragada: belt, you have the cameras captured and that feed the data directly into some systems. What happens if a major issue is found like a there's a crack on the sheet or something? Naveen Puttagunta: Another Satyasri Prabhakar Mantripragada: Uh does the production halt or they take it over and then they just cut that piece and then assemble them again or what happens next? Are there any Vijay Kamineni: Uh I'll give I'll give you a I'll give you Satyasri Prabhakar Mantripragada: alerts? Vijay Kamineni: a consultant answer. It depends. So no, you didn't want to hear that. Um so if it's a continuous sheet if it's a continuous sheet Naveen Puttagunta: Heat. Vijay Kamineni: uh there's a continuous monitoring and they tag the sheet and so sometimes they give a discount to the client. It'll all go depending on where it if it's in the head or tail part of it they can easily cut it. 00:22:13 Vijay Kamineni: If it's right in the middle they can cut it and make it into two coils. Naveen Puttagunta: Heat. Vijay Kamineni: But if it is three quarters into the sheet, Naveen Puttagunta: Heat. Vijay Kamineni: they'll probably give a discount for the last three quarters to the client. Uh some defects, you can't send some defects. Naveen Puttagunta: Heat. Vijay Kamineni: It's a it's a um pin holes. If the drink is going to leak from it, they won't ship it. Right. So the that's a that's a you're you're thinking the right way but it's Satyasri Prabhakar Mantripragada: Yep. Naveen Puttagunta: Heat. Vijay Kamineni: each plant same plant for different customers is different. So uh they divert the product. So if a particular defect comes one customer can't deal with it they'll ask for Naveen Puttagunta: Heat. Vijay Kamineni: a for a refund but other customer can't even find it their machines will run Naveen Puttagunta: Heat. Vijay Kamineni: it. So I'm making one for you. I find a defect. Naveen Puttagunta: Heat. Vijay Kamineni: Oh, Naven's machine can't find it. 00:23:06 Vijay Kamineni: So, I'll I'll transfer that to Naven and make a new coil for you. So, there's a lot of that you can't get. Yeah. Satyasri Prabhakar Mantripragada: And uh do you also get insights or does somebody require insights related to the preventive maintenance of machines versus the scheduled Vijay Kamineni: Oh yeah. Satyasri Prabhakar Mantripragada: maint? Vijay Kamineni: Oh yeah. I mean uh 85,000 Logan aluminum has 85,000 tags per second per millisecond. IoT data there is 85,000 tags uh being captured at a millisecond interval. So yes, Naveen Puttagunta: Wow. Vijay Kamineni: and it's like it's like a car wash, right? So it uh uh it's it's a very complex process. So one one machine has three stands and it's taking a continuous sheet and making it thinner on every pass, right? Naveen Puttagunta: It's Vijay Kamineni: Every every every stand it goes through one stand it becomes half the thickness. Naveen Puttagunta: beautiful. Vijay Kamineni: It goes through the next and the next stand it half the thickness. That means those three stands three stand should run at different speeds because every 00:24:21 Naveen Puttagunta: Don't Vijay Kamineni: time the thickness reduces length is increasing. So the last stand should run three times faster than the first stand and they all should have they all should dance properly so they won't tear the sheet. Naveen Puttagunta: Right. Vijay Kamineni: So it's fun. It's we can get into that. Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: Stamping is completely different from rolling. Uh stamping is you stamp and take the pieces. It's a very discrete process. It's not a continuous process. Naveen Puttagunta: Good Vijay Kamineni: Slitting is a discrete process, not a continuous process. Rolling is a continuous process. Naveen Puttagunta: time. Vijay Kamineni: Melting is a batch process. You just you have a pot or you don't have a pot. There's no there's no separation in a melting pot. and two 250,000 lbs per part. So this the chemical analysis at that time and those kind of Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: things. Satyasri Prabhakar Mantripragada: And in case of individuals um we Naveen Puttagunta: Thank you. Satyasri Prabhakar Mantripragada: can Naveen Puttagunta: I'll I'll put it. 00:25:31 Naveen Puttagunta: No. Vijay Kamineni: Sorry, I didn't hear Satyasri Prabhakar Mantripragada: Yeah, sorry. Vijay Kamineni: question. Satyasri Prabhakar Mantripragada: In case of uh individual sheets, metal sheets, uh there can be some kind of tagging or numbering. Uh but how do we distinguish the a particular position on a coil? For example, Vijay Kamineni: length number and length. Satyasri Prabhakar Mantripragada: coil do. Vijay Kamineni: K number length and Naveen Puttagunta: Okay. Vijay Kamineni: width. Satyasri Prabhakar Mantripragada: Okay. Do we get the relative distance from the start of the coil uh to Vijay Kamineni: Yeah. Naveen Puttagunta: Okay. Vijay Kamineni: Yeah. They they'll they'll have a relative distance and uh they have the position. Satyasri Prabhakar Mantripragada: Okay. Vijay Kamineni: So we actually show the entire um 3,000 ft sheet as an image and show all the defects in a visual by location by distance from the head and distance from the edge. Satyasri Prabhakar Mantripragada: Mhm. Vijay Kamineni: So there's a operator side edge and uh so but we're getting into so much details that's how one machine works. Naveen Puttagunta: stupid. Vijay Kamineni: If you if you're if you're interested, I can tell you some more things. 00:26:38 Vijay Kamineni: So, coil spools and unspools, Naveen Puttagunta: She's Vijay Kamineni: right? You you unspool, run, do some work on it, and spool it again, right? Typically, Satyasri Prabhakar Mantripragada: head. Vijay Kamineni: so if you unspool it from the top and spool it from the top to bottom, then the top stays on the top. Satyasri Prabhakar Mantripragada: Hey Vijay Kamineni: Bottom underside of the sheet stays underside. Naveen Puttagunta: ready. Vijay Kamineni: If you take from the top and roll it down, the top becomes outside becomes inside, inside becomes outside. Satyasri Prabhakar Mantripragada: Got it. Got Vijay Kamineni: So if you had a spot on the outside, Satyasri Prabhakar Mantripragada: it. Vijay Kamineni: depending on whether you do under or over, it either goes up or down. So um if you get that detail, they'll they'll eat us up on any assumption we make because now we got too close into their process. Naveen Puttagunta: That's Satyasri Prabhakar Mantripragada: Yeah, we Vijay Kamineni: as much as they know and any problem we say they'll say either we solved it or you or you Satyasri Prabhakar Mantripragada: becoming 00:27:38 Vijay Kamineni: can't solve it even if you say you'll solve it because they tried and they could not solve it so I would stay away from those and what they can't do is this this thing they Naveen Puttagunta: Perfect. Vijay Kamineni: can't ask if they have a set of data they can't answer questions as fast as we can without knowing the problem they have the knowledge of the problem we can Naveen Puttagunta: Heat. Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: answer questions on unknown problems. Naveen Puttagunta: Heat. Vijay Kamineni: So what they don't know we can get like right now I don't know what the data is but Satyasri Prabhakar Mantripragada: Okay. Vijay Kamineni: now they have to do a traditional way they have to do a SPC analysis for standard deviation and all that they have to use stats I just ask my brain what's my stats now I have a standard deviation and a range then I can go and find everything that's over 7.15 as Naveen Puttagunta: It's been Vijay Kamineni: outliers and get my what is how many of them are in my top quartile and start working on that. Naveen Puttagunta: time. 00:28:36 Vijay Kamineni: So we should stay in that space. Satyasri Prabhakar Mantripragada: Yeah. Got it. Vijay Kamineni: Um good questions. Uh but it get like you say it gets complicated as you go deeper. Satyasri Prabhakar Mantripragada: Now basically why the point of asking these questions is to understand to what extent should we create the synthetic data when should we stop? Vijay Kamineni: So the answer to that is you have to have a story to tell. Naveen Puttagunta: I Vijay Kamineni: You make up the story first and create the data for that. Naveen Puttagunta: Okay. Vijay Kamineni: So I'm asking I I asked the standard deviation and range question on the screen here you can Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: see because I'm hoping to see this insight. This process has high variability. If you have a high variability, you can't control the it's a it's out of control process. Satyasri Prabhakar Mantripragada: Wow. Vijay Kamineni: So now SPC will come and say I need to get my process into control. I need to get the range closer before I make a lot of corrections to so those are the things that they would know but you have it already. 00:29:50 Vijay Kamineni: Enterprise brand can do all of that. You have to tell that story. Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: that I I don't know if I'm making sense Naveen Puttagunta: No, Vijay Kamineni: there. Naveen Puttagunta: no. Satyasri Prabhakar Mantripragada: Oh. Naveen Puttagunta: I I I get uh I get it completely simma. So netn net I think uh it's not a standard process. It's not a there is no there is no such thing as a standard. So it doesn't matter what we are embellishing with because anyway you make up that story and Vijay Kamineni: Correct. Naveen Puttagunta: then uh embellish it with data that fits that story. Vijay Kamineni: Correct. Naveen Puttagunta: So I would say um I would say uh prabaka just to give variations right let's embellish it let's tag it with um two stages in each stage there is a machine and a person and there is a material and material comes from let's say three different vendors now we will you know we will simulate the data such that you know during a certain time period one particular vendor's uh material is is giving one particular type of error obviously you know it won't work if you do everything randomly because then there is no pattern that you can uh tell a story about right so 00:31:10 Vijay Kamineni: Right. Naveen Puttagunta: we and again you know I will leave it to Pratima and Dia and Sima and others to Vijay Kamineni: Right. Naveen Puttagunta: tell the story that look you know we understand this is a little bit of simulated data but at the same time this is what the capability of what we can do. So whatever your data is, we're able to find such insights. Satyasri Prabhakar Mantripragada: got it. Vijay Kamineni: Right. Naveen Puttagunta: So so that's what I would suggest praar let's let's simulate two stages. Satyasri Prabhakar Mantripragada: Sure. Naveen Puttagunta: Each stage has a machine and a team member and uh you know uh and you know material comes from uh one particular vendor. That's what we'll tag the data with the 5,000 images that we have and then we'll we'll we'll train enterprise rate. Don't give a separate instance. Attach it to the same uh database. Whatever Simma is playing with right now, Satyasri Prabhakar Mantripragada: Okay. Vijay Kamineni: Right. Naveen Puttagunta: right? Put put it in the same thing because it's just it's just one more table, right? 00:32:10 Naveen Puttagunta: So it doesn't need another instance. Satyasri Prabhakar Mantripragada: Okay. Naveen Puttagunta: If you ask one type of questions, it'll go after one type of data. If you ask other type of questions, it'll go to the other data. That's it. Vijay Kamineni: Yeah. Satyasri Prabhakar Mantripragada: Okay. Uh Rasher, is that Vijay Kamineni: Yeah. Praash look at this answer. This is probably uh highlight of enterprise brain Satyasri Prabhakar Mantripragada: clear? Rajashekar G: Yes, Vijay Kamineni: strength. Rajashekar G: sir. Vijay Kamineni: The question is see how simple the question is. What are the standard deviation and range of defects? Right? It's a very simple question. Look at the kind of information enterprise brain gave. It said it may be have extremely high variability. Naveen Puttagunta: Here Vijay Kamineni: So I need to do something with it. uh we should expect a bell curve the normal distribution so it's very flat Naveen Puttagunta: we Vijay Kamineni: so your we have our quality control triggers might be finding randomly Naveen Puttagunta: go. Vijay Kamineni: so it identified it's a random data that's a nice 00:33:10 Rajashekar G: That's me. Vijay Kamineni: and uh it said your 6G network latency may be called causing false Satyasri Prabhakar Mantripragada: Mhm. Vijay Kamineni: defects because they should not have so many it's a so that's What manufacturers will go and hunt now and say oh I need to fix these things I mean one small question look at how much information it gave to actionable information to the Naveen Puttagunta: Heat. Vijay Kamineni: manu that's the story enterprise brain has that strength no human can do it in that Naveen Puttagunta: Heat. Vijay Kamineni: in the time I did with data that I don't Satyasri Prabhakar Mantripragada: Yeah. Vijay Kamineni: know and I don't even know whose factory it I asked a simple question. Satyasri Prabhakar Mantripragada: Okay. Vijay Kamineni: Now I can go and tell them, okay, you need to look at your 6G network latency. That's the beauty of what we have and what the manufacturing won't Naveen Puttagunta: I'll pick that Vijay Kamineni: know. Naveen Puttagunta: I come. Okay. So, uh I need to drop off. Vijay Kamineni: Yeah. Naveen Puttagunta: I need to get started to the airport. 00:34:31 Naveen Puttagunta: So, thanks so much. Vijay Kamineni: No, you're welcome. Naveen Puttagunta: So, Vijay Kamineni: This is exciting. Naveen Puttagunta: yeah. Yeah, it is very exciting. So Prabhakar Rajar I hope uh you know that that's how let's tag the data Satyasri Prabhakar Mantripragada: Yeah. Naveen Puttagunta: then put it in the same database so that it can give you know maybe it'll give you some we we'll see how it behaves if it behaves too weirdly because we put Rajashekar G: Sure, Naveen Puttagunta: this two data pieces in the same one then maybe we'll separate it out Rajashekar G: sir. Vijay Kamineni: All Naveen Puttagunta: okay thank you and I'll I'll talk to you Vijay Kamineni: right. Do Yeah. Looking forward to it. You gave me a You gave me a Disney World. I'll be playing with this a lot. He gave me a toy. I can't get any work. See you. See you. Bye. Satyasri Prabhakar Mantripragada: Yeah. Uh if uh I'm not sure if we are able to capture the uh prompts uh so that we can understand where things are going wrong if at all there is any uh the history 00:35:33 Vijay Kamineni: Yeah, it'll be in the chat history, Satyasri Prabhakar Mantripragada: is okay Rajashekar G: We can we can we can take it sir. Vijay Kamineni: right? It'll be in the chat history. Rajashekar G: We can login with this. Satyasri Prabhakar Mantripragada: fine. Vijay Kamineni: You can go and look at what I'm doing and read read those and uh we'll Rajashekar G: Yeah. Vijay Kamineni: we'll see. Satyasri Prabhakar Mantripragada: Yeah. Yeah. Vijay Kamineni: So, so not that not the fact that it is written the data. Satyasri Prabhakar Mantripragada: Sure. Vijay Kamineni: This is more interesting. You you all should pay attention to this. So what I will do here is um um it said false defect Satyasri Prabhakar Mantripragada: Correct. Vijay Kamineni: triggers in the uh automatic UR cameras right defects in uh 6G latency Rajashekar G: Those Vijay Kamineni: Okay. Satyasri Prabhakar Mantripragada: Huh? Vijay Kamineni: So, what do we do? Satyasri Prabhakar Mantripragada: Uh that's pro uh Rashak we'll have to do something about giving an indication it's Vijay Kamineni: Time out. Satyasri Prabhakar Mantripragada: processing. No it's not maybe it is Vijay Kamineni: Oh. Satyasri Prabhakar Mantripragada: processing for Rajashekar G: are in progress. Vijay Kamineni: Sounds Rajashekar G: Yeah. Vijay Kamineni: good. Satyasri Prabhakar Mantripragada: enterprise brain we we have given that indication that it's processing doing something something something uh but this layer has come recently on top of visual defect detection. So we just have to get those things here. Vijay Kamineni: The correlation between latency and defix is non non-existing. Packet loss shows no meaningful impact on error rate. So those two are there's no correlation. Satyasri Prabhakar Mantripragada: Sure. Vijay Kamineni: So I don't have to deal with them right now. Um so I'll I'll go look at that and I'll ask for follow-up questions. So you'll see that. Let me play with it. I'll I'll give you all uh feedback. Rajashekar G: Thank Satyasri Prabhakar Mantripragada: Thank you. Okay. Stop Transcription ended after 00:37:50 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.