Feb 26, 2026 Connect with Vijay - Transcript 00:00:00 Prathima Inolu: Okay. Uh, Salesforce I have a call with Sation a bit. Naveen Puttagunta: Okay. Sure. Ratma. Prathima Inolu: Bye. Yeshwanth Reddy Yerraguntla: What the Naveen Puttagunta: So um what I was saying uh guys is see we you know um Yeshwanth Reddy Yerraguntla: hell? Naveen Puttagunta: some and I I thought data sets meant probably like multiple data data right like multiple tables and all of that but it looks like most of them is just like one single CSV which means that it's just one uh table which is not so um demonstrative. Right. But if you take okay quality, Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: reliability, safety, if there are some things that are multiple, you know, they combine different aspects th those would be ideal. So can we look at something that uh can can give us more of those um and then you know demonstrate enterprise brain on that. Yeshwanth Reddy Yerraguntla: He was at the end saying that uh you can join these tables multiple tables into one table using fake ids. Naveen Puttagunta: Sure. Yeshwanth Reddy Yerraguntla: You can we can try that. 00:01:17 Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: I'm not saying that's the only way. We'll definitely look into other things but uh even if uh Rajashekar G: You mean to say multiple data sets Yeshwanth Reddy Yerraguntla: sorry Rajashekar G: into one data set or they Yeshwanth Reddy Yerraguntla: correct correct correct yeah so zia as a whoever is the uploader they uploaded multiple of these versions of the data set smart manufacturing multi-agent control data set intelligence manufacturing data set IoT integrated predictive Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: maintenance data set. We have to study understand each of these data set is trying to figure out or trying to point to one aspect of whatever we just discussed right quality, reliability, safety, productivity. For example, intelligent manufacturing data set is talking about efficiency status. Efficiency translates to uh productivity. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Allah if I'm not wrong if there are four different data sets talking about four different aspects how can we combine all of them into one single uh uh source is what we have to uh try to look Naveen Puttagunta: Yeah. 00:02:39 Yeshwanth Reddy Yerraguntla: at m Naveen Puttagunta: And just I just want to make sure that you know we are not showing something too trivial. Yeshwanth Reddy Yerraguntla: I still feel we are we will be showing something trivial only. Naveen Puttagunta: So then we can we can then show that to Vijay get his opinion and in fact if we are able to quickly uh you know do that and say okay can can you can ask him for the real business questions right because our questions will be too trivial. So he can give real business questions and we can say okay can we showcase that uh that things work. Got Yeshwanth Reddy Yerraguntla: Right. Yeah. Naveen Puttagunta: it. Yeshwanth Reddy Yerraguntla: But okay. So we have to first of all I think we can use chat jupit and guess what could those business questions be based on these data sets. Naveen Puttagunta: Okay. Fantastic. Yeshwanth Reddy Yerraguntla: JC um yeah given that this is a single table. Rajashekar G: I'm Yeshwanth Reddy Yerraguntla: Okay. Time stamp. I see time stamp itself can become the ID on 00:04:16 Rajashekar G: a time stamp is there and different parts also there machine ID allow If you said two two data sets but time stamps match Naveen Puttagunta: Okay, Yashwant Rashikar, I'm you know I am dropping off because I'm going to work with Ward on something but you got the idea covered. Rajashekar G: one Yeshwanth Reddy Yerraguntla: Okay, we'll Naveen Puttagunta: Please go through and you know we don't have to uh I mean Yeshwanth Reddy Yerraguntla: continue. Naveen Puttagunta: we can go back to Vijay no worries we can go back to him with some of proposal okay this is what we're doing and get an okay and move forward we can go to him when we actually put it in into this and then say hey this is what it looks like uh is it reasonably uh exciting enough for somebody to get an idea of oh this is what is possible and And you know given real data we can do a lot more. Yes. Yes. Prabhaka. Satyasri Prabhakar Mantripragada: uh not related to your point and just uh just checking on the schedule of the tomorrow's meeting 9:30 or 9 9:00 a.m. 9:30 to 10:30. 00:05:20 Naveen Puttagunta: See Satyasri Prabhakar Mantripragada: Does it work? Vijay though. Vijay though. Naveen Puttagunta: I in fact if we are going to do this if we nail down then it is worth taking his time tomorrow morning. Satyasri Prabhakar Mantripragada: Mhm. Naveen Puttagunta: If we need time to then identify or and then come to a conclusion of okay let's take these four data sets Satyasri Prabhakar Mantripragada: Correct. Naveen Puttagunta: let's join this here then we can have a very quick 10 minute to show that okay then Satyasri Prabhakar Mantripragada: So tomorrow morning we'll decide at what time based on H1 Raja's progress. Naveen Puttagunta: yes okay thank you guys I'm dropping Satyasri Prabhakar Mantripragada: Okay. Fine. Anyway, sure. Sure. Naveen Puttagunta: off so quality reliability safety and uh Satyasri Prabhakar Mantripragada: Yeah. One more Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: thing. Yeshwanth Reddy Yerraguntla: Efficiency, Naveen Puttagunta: uh uh efficient productivity right E4 factors Yeshwanth Reddy Yerraguntla: productivity. Naveen Puttagunta: two out of four three out of four what he was saying is safety because in any given plant you know safety issues so 00:06:20 Satyasri Prabhakar Mantripragada: Sure. Naveen Puttagunta: that won't be impressive because of course fake but then it look like there's an accident happening every day so you know Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Thank you Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: Yeah. Rajashekar G: sir. Satyasri Prabhakar Mantripragada: In ask that should we record the call or not? Rajashekar G: Regarding uh stop. Satyasri Prabhakar Mantripragada: You can check. Rajashekar G: Okay. Session ended after 00:06:56 00:19:50 Yeshwanth Reddy Yerraguntla: correct. I agree. Oh, no. Rajashekar G: model. Yeshwanth Reddy Yerraguntla: Oh, Rajashekar G: This is all synthetic. Yeshwanth Reddy Yerraguntla: no. But are you saying because the customers will upload their Rajashekar G: So imagine Yeshwanth Reddy Yerraguntla: own images something is going to go wrong Rajashekar G: Like if they upload their images, it doesn't work. Yeshwanth Reddy Yerraguntla: if they upload their images is where I'm saying we will not even allow them to do that. Why will why will it not going to happen? So what's the uh fear? Rajashekar G: Uh but my point is let's Yeshwanth Reddy Yerraguntla: H Rajashekar G: say some 00:20:48 Yeshwanth Reddy Yerraguntla: okay. Rajashekar G: images are there. Let's say this image is here this image is here root cause but im Okay. Yeshwanth Reddy Yerraguntla: Oh, Rajashekar G: Everything Yeshwanth Reddy Yerraguntla: one second. One second. So, Rajashekar G: is Yeshwanth Reddy Yerraguntla: what we might show on the screen manufacturing people will say this is not the root cause. Rajashekar G: maybe. Yeshwanth Reddy Yerraguntla: Is that what you're saying? Rajashekar G: Yeah. Yes. Yeshwanth Reddy Yerraguntla: Is there anything additionally problematic on top of it? No, I know it's simulated. Rajashekar G: What? Yeshwanth Reddy Yerraguntla: Hence whatever reasons we might give is bogus. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Any other issue? Rajashekar G: No issues. That is the major concern from my side. And what the expectation is. So reliability, Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: quality, these kind of Yeshwanth Reddy Yerraguntla: Hm. Rajashekar G: question the different right? Yeshwanth Reddy Yerraguntla: Um Rajashekar G: Whatever the data set he was showing that is a little different from this one whatever we have Yeshwanth Reddy Yerraguntla: H that is different. 00:22:08 Rajashekar G: prepared this Yeshwanth Reddy Yerraguntla: I agree. Rajashekar G: is like some efficiency status it is showing error rate it is showing uh let's say some question is like this uh what what's causing my scrap Yeshwanth Reddy Yerraguntla: H. Correct. Rajashekar G: machine let's say which machine is causing me more errors and what is the least in error rate what is the maximum in error rate and that one mapped with which machine ID as simple as Yeshwanth Reddy Yerraguntla: H but you will have Rajashekar G: that But are any Yeshwanth Reddy Yerraguntla: multiple rows with bad error Rajashekar G: questions? Yeshwanth Reddy Yerraguntla: rate. Rajashekar G: H. Yeshwanth Reddy Yerraguntla: I mean yeah I mean I I see what you are saying. Rajashekar G: Oh Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: no. Yeshwanth Reddy Yerraguntla: Get those 50 rows and simply say these 50 rows are the culprits. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Um, repeat. Rajashekar G: I mean machine ID is repeat. Let's club Yeshwanth Reddy Yerraguntla: What is causing what is Rajashekar G: and what he was multiple. Yeshwanth Reddy Yerraguntla: Yeah. He was not taking it in a very serious fashion like saying 00:23:56 Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: which is why I was also thinking Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: the simulations. So that if I show an image and show the root cause and both are not matching, I'll be I'll not be taken Rajashekar G: H Yeshwanth Reddy Yerraguntla: seriously. But this is all simulation Rajashekar G: right. Yeshwanth Reddy Yerraguntla: data. Um okay we have those four defect types right ABC D in each one of them what is the root Rajashekar G: H Yeshwanth Reddy Yerraguntla: person. Rajashekar G: four image four root cause uh four what is the root cause it may suggest few Thanks. I love Yeshwanth Reddy Yerraguntla: So to some extent uh it might show reasonably Rajashekar G: you. Yeshwanth Reddy Yerraguntla: okay root causes up problem. uh Radhu see end of the day they want to enable conversation on data set that has vision that also supports showing evidence Rajashekar G: Right. But evidence means like Yeshwanth Reddy Yerraguntla: evidence is uh correct Rajashekar G: but I don't know like so what was the process they thinking is there will be some data sets and NLP to SQL and Not uh this is something different like vision defect detection process is something different that's 00:26:20 Yeshwanth Reddy Yerraguntla: Real time simulator real time Rajashekar G: it Yeshwanth Reddy Yerraguntla: aspect. Rajashekar G: is different like vision defect that is completely different I think allows Which kind of representation or visual representation root Yeshwanth Reddy Yerraguntla: H Rajashekar G: predictive maintenance score error rate efficiency status So They just Yeshwanth Reddy Yerraguntla: Again asking these questions and getting back answers is one thing but can it surface the same intelligence by itself Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: enterprise dashboard it's like automatically showing intelligence is not waiting for me to ask questions. Rajashekar G: Direct. It might give me the insights. Yeshwanth Reddy Yerraguntla: Uh yeah to to achieve that it's not so difficult decision build but these will be sort of static. Rajashekar G: Right. Right. study ideas. Yeshwanth Reddy Yerraguntla: response. But to ask questions, I want to show intelligence automatically. Currently we are only showing NLSQL components. Right? Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Is there an optimal temperature vibration range? Rajashekar G: Okay. Answers means representation of 00:32:02 Yeshwanth Reddy Yerraguntla: uh representation. Rajashekar G: light. Yeshwanth Reddy Yerraguntla: So based on this data set, how what model is used for getting answers? Rajashekar G: Okay. I'm done. Sure. Yeshwanth Reddy Yerraguntla: The age of which machines have highest defect rate. Rajashekar G: You'll need a layered modeling stack depending on the question type. Discuss the questions what's happening. No ML model is required. SQL pandas aggregations diagnostic. Why this is happening? Correlation persona spaceman regression models predictive models. Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: Okay, then fine. Rajashekar G: Yeah, sir. Satyasri Prabhakar Mantripragada: 8:30. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Okay. Okay. Continue. Rajashekar G: link Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: out. Yeshwanth Reddy Yerraguntla: Um yeah, purpose based on every task you use a different model. Rajashekar G: and then directly like it's like NL2SQL kind of thing with some intelligence Yeshwanth Reddy Yerraguntla: First thing basic on top of that we have to build some uh models up front and uh questions the uh call it should be in a position to give answers. 00:35:33 Rajashekar G: Good night. models like ML models. Yeshwanth Reddy Yerraguntla: One or two. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Cast one Rajashekar G: Clarity. Yeshwanth Reddy Yerraguntla: person. Rajashekar G: So previously what we are doing multiple tables it has a schema based on that it is understanding there will be multiple Yeshwanth Reddy Yerraguntla: Oh no. Yeah. Rajashekar G: columns. Yeshwanth Reddy Yerraguntla: Oh no. which is where ML models will Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: come. Rajashekar G: fine. Yeshwanth Reddy Yerraguntla: Yeah. Are we dealing with only one this one data set? We have to Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: discuss Rajashekar G: Or else once we are uh at certain point we can ask for are we in the right direction or not. Yeshwanth Reddy Yerraguntla: at least one more data set Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: and so let's try Rajashekar G: Help. This one. Yeshwanth Reddy Yerraguntla: H RT agent control data Rajashekar G: Task execution, learning efficions and realtime control agent ID agent. Yeshwanth Reddy Yerraguntla: Multi- aent data set agent is machine on 00:38:09 Rajashekar G: Oh no. Okay. Let's say this one are doing in inspection. This one is doing welding execution time, Q value, Yeshwanth Reddy Yerraguntla: H Rajashekar G: machine usage, energy consumption, production efficiency, security event for Yeshwanth Reddy Yerraguntla: basically data set. Rajashekar G: ZP Yeshwanth Reddy Yerraguntla: No, but connect our Rajashekar G: H common layer. Yeshwanth Reddy Yerraguntla: easy Rajashekar G: Or else what we can do is agent column that can be possible mission operation. Okay, that's not Directly implementation System efficiency Yeshwanth Reddy Yerraguntla: H Rajashekar G: both are like same complexity. Let's Yeshwanth Reddy Yerraguntla: Fake columns. Rajashekar G: say detection this P adjustment security Yeshwanth Reddy Yerraguntla: H Rajashekar G: event security is data breach DD attack unauthorized Did it die? Yeshwanth Reddy Yerraguntla: See first targets focus first data set focus second data Rajashekar G: Efficiency Yeshwanth Reddy Yerraguntla: set Rajashekar G: efficiency. I think what is the output? Yeshwanth Reddy Yerraguntla: targets different. You can Rajashekar G: Club. Yeshwanth Reddy Yerraguntla: uh Rajashekar G: Okay. Predictive maintenance. 00:41:25 Yeshwanth Reddy Yerraguntla: Predictive maintenance generally very hot topic manufacturing law machine health. Rajashekar G: state healthy or faulty two classes vibration acostic Yeshwanth Reddy Yerraguntla: M Rajashekar G: temperature current IMF123 what are these intrinsic modern functions decomposing component representing underlying patterns sensor data. These are all some IoT information. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Let's say Multiple things ID operational temperature. Both are almost same. Yeshwanth Reddy Yerraguntla: I thought they got audio. Rajashekar G: Next. I sensor data for predictive maintenance. Smart manufacturing energy metrics which is related prediction via classifier chain time stamp machine logs capturing operations over convention and work matrix for engineer. Hello. Good day. Most basically machine Yeshwanth Reddy Yerraguntla: Got Rajashekar G: operation. Yeshwanth Reddy Yerraguntla: to Rajashekar G: What is the vibration? What is the temperature? efficiency Yeshwanth Reddy Yerraguntla: Yeah. Got her. Rajashekar G: almost but we might be think like we can link with something else let's say raater material operator can link and try 00:45:28 Yeshwanth Reddy Yerraguntla: a Rajashekar G: talk. The current data set is giving me the machine information and the efficiency. So I'd like to link with the link this data set with some other data sets also to give a dashboard kind of information on the entire uh productivity and efficiency calculations. So what can I link this data set with? Satyasri Prabhakar Mantripragada: Flickering. Yeshwanth Reddy Yerraguntla: Can you say Rajashekar G: fix. Yeshwanth Reddy Yerraguntla: that? Rajashekar G: Okay. Analyze the current sales plan. Map it to the business target. Pro the forecast for the future and highlight the risk. So just direction changes. Satyasri Prabhakar Mantripragada: Hello. Different MD Yeshwanth Reddy Yerraguntla: Mark, don't Satyasri Prabhakar Mantripragada: style. Rajashekar G: Mark downstairs. Yeshwanth Reddy Yerraguntla: sell. Satyasri Prabhakar Mantripragada: But huh. Yeshwanth Reddy Yerraguntla: First thing So Rajashekar G: Ra. Yeshwanth Reddy Yerraguntla: download simplify. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Mhm. Rajashekar G: Huh? We are still looking. Satyasri Prabhakar Mantripragada: X-axis could correct. Mhm. Rajashekar G: I thought Satyasri Prabhakar Mantripragada: Okay. 00:49:07 Rajashekar G: better. Satyasri Prabhakar Mantripragada: Okay. Download. Okay. I'll try another question then. Okay. Yeah. Yeshwanth Reddy Yerraguntla: answer is Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: solid. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: What should what should be the solution? All Rajashekar G: It's always link is in the Yeshwanth Reddy Yerraguntla: right, Rajashekar G: color. Okay. Satyasri Prabhakar Mantripragada: JSON file. Yeshwanth Reddy Yerraguntla: check Rajashekar G: Yes. Uh Germany 3 is working better. Yeshwanth Reddy Yerraguntla: 2.5 is too old for these things. Rajashekar G: So what I'm saying is Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: model UI changes everything Yeshwanth Reddy Yerraguntla: Got it. Rajashekar G: second. Satyasri Prabhakar Mantripragada: Basically Rajashekar G: questions. Satyasri Prabhakar Mantripragada: question Rajashekar G: Okay. Okay, sir. Satyasri Prabhakar Mantripragada: scope Salesforce Jira drives only even if it gives incorrect Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: results that's fine but at least Rajashekar G: Okay. Okay. Now it has given me some points. Yeshwanth Reddy Yerraguntla: Ch Rajashekar G: I am sharing 00:51:52 Yeshwanth Reddy Yerraguntla: some Rajashekar G: screen share. Yeshwanth Reddy Yerraguntla: better. Rajashekar G: Okay. condition. Yeshwanth Reddy Yerraguntla: Got a lot. What Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: should Rajashekar G: So current system what other should I integrate to get true end to end productivity efficiency dashboard ERP production planning data work orers production target product type batch information was expected bill of materials which product causes highest defect which stresses machines small work force shift data. Same. Yeshwanth Reddy Yerraguntla: Yeah, but in simulate if I had to prioritize step one ERP Rajashekar G: Oh, Yeshwanth Reddy Yerraguntla: plus Rajashekar G: ERP plus financial step two maintenance logs. Step three, workforce supply chain 80% business visibility. Now, let's think big. building a dashboard sensors models cost layer a explanation C6 Shall we try for other data set Yeshwanth Reddy Yerraguntla: Huh? Rajashekar G: remaining let's say raw material renders Fine. I know. Yeshwanth Reddy Yerraguntla: already. They got it. Rajashekar G: machine extra columns like temperature Yeshwanth Reddy Yerraguntla: Yeah. 00:55:58 Rajashekar G: images. Yeshwanth Reddy Yerraguntla: We get the data set Rajashekar G: Why images? Any Yeshwanth Reddy Yerraguntla: just because we have we can show by Rajashekar G: question Yeshwanth Reddy Yerraguntla: images. Fair enough. Rajashekar G: I map? Okay. question images. Yeshwanth Reddy Yerraguntla: Yeah, makes sense. Rajashekar G: Mhm. Uh at least first things like questions. Oh, Yeshwanth Reddy Yerraguntla: download. Rajashekar G: question Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: that. Okay, this is a good direction. Satyasri Prabhakar Mantripragada: Check Haggle. Uh, Rajashekar G: machine information. Satyasri Prabhakar Mantripragada: okay. Rajashekar G: What are the vibration? What is it getting? So questions it has given us lot of Satyasri Prabhakar Mantripragada: Okay. Rajashekar G: questions. Satyasri Prabhakar Mantripragada: predictive maintenance. If vibration is too high, it is highly likely that it will produce defective sheets. Is that Rajashekar G: Oh no. Oh Satyasri Prabhakar Mantripragada: so? Rajashekar G: no. Let's say which machine produce highest units per hour? Production speed units per hour. 00:59:05 Rajashekar G: What is average production speed per efficiency status? questions. Satyasri Prabhakar Mantripragada: This can also be the these can also be the Rajashekar G: Questions. Satyasri Prabhakar Mantripragada: questions based on Which role is using the system? Questions may head of the organization may only look at across plants which plant is yielding more revenue stand. So Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: roll Rajashekar G: Mission let's plan it as column plants and see plant column one to five for each one few Satyasri Prabhakar Mantripragada: possible, huh? Rajashekar G: machines. Satyasri Prabhakar Mantripragada: and shift score. Rajashekar G: Oh no. Yeshwanth Reddy Yerraguntla: All right. Satyasri Prabhakar Mantripragada: shift Rajashekar G: Day shift Yeshwanth Reddy Yerraguntla: I don't know. Rajashekar G: night. Satyasri Prabhakar Mantripragada: down 6 hours down Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: between two shifts. There is a down time between three shifts. After three shifts, there will be one down time. Yeshwanth Reddy Yerraguntla: Maybe. Yeah. Satyasri Prabhakar Mantripragada: If we go too deep into data then customers they rather than focusing on the 01:01:18 Rajashekar G: Right. Satyasri Prabhakar Mantripragada: solution they'll try to find mistakes in our understanding. For example, if you measure something in say square meters, Yeshwanth Reddy Yerraguntla: H. Satyasri Prabhakar Mantripragada: they may say see who asked you to measure in square meters. Rajashekar G: H Satyasri Prabhakar Mantripragada: We our standard protocol is to measure in square foot. Yeshwanth Reddy Yerraguntla: And we can't tell the client that this is all simulated data, Satyasri Prabhakar Mantripragada: Uh maybe we can Yeshwanth Reddy Yerraguntla: right? Satyasri Prabhakar Mantripragada: um so tomorrow we'll present the same thing. So if at all they get a chance to access their Yeshwanth Reddy Yerraguntla: Yeah, Satyasri Prabhakar Mantripragada: data is Yeshwanth Reddy Yerraguntla: better option. Satyasri Prabhakar Mantripragada: the next they can play with the system to see what they look for. Are they getting it or not? Yeshwanth Reddy Yerraguntla: I Check character. Rajashekar G: Can valid or shall we proceed? Satyasri Prabhakar Mantripragada: The visual detection starting actual concept. Um what was the diagnostic layer decision making? Decision making good. So you don't just present a dashboard that this is where we stand, this is where we stand, this is the output. 01:03:26 Satyasri Prabhakar Mantripragada: If you are not able to produce a certain kind of output to a certain volume, what is causing this? What actions will prevent this? Rajashekar G: Got it Let's say based on this efficiency unit then we can Okay. Satyasri Prabhakar Mantripragada: the expected fine. Rajashekar G: Uh Satyasri Prabhakar Mantripragada: What I was expecting in that case is expected yield versus actual yield. Rajashekar G: oh. service. Satyasri Prabhakar Mantripragada: Can we discuss 8:30 discussion what she was mentioning is if you could figure it out uh machine Rajashekar G: All Satyasri Prabhakar Mantripragada: A in shift 2 by operator C is causing more Rajashekar G: right. Satyasri Prabhakar Mantripragada: issues or every Wednesday second shift there are more issues. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Uh scenario. Just Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: accept Rajashekar G: And one more thing. Satyasri Prabhakar Mantripragada: masculine colors, Rajashekar G: Oh, Satyasri Prabhakar Mantripragada: feminine colors. Yeshwanth Reddy Yerraguntla: First time I know. Rajashekar G: right. Satyasri Prabhakar Mantripragada: She was suggesting purple pink shades more feminine Rajashekar G: System. Satyasri Prabhakar Mantripragada: approach whereas these sheet industries and all they represent 01:05:51 Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: power. So more more masculine darker Rajashekar G: Thank my Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: colors Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So right now for Okay. Satyasri Prabhakar Mantripragada: If we have to change a certain styling Rajashekar G: Uh, Satyasri Prabhakar Mantripragada: Okay. And CSS Rajashekar G: I put Global level Satyasri Prabhakar Mantripragada: then fine. Okay. Rajashekar G: CSS. Satyasri Prabhakar Mantripragada: What takes more effort? that aligns with anyways enterprise brain or Rajashekar G: Right. Satyasri Prabhakar Mantripragada: so you figure it out Rajashekar G: Data set is the major Satyasri Prabhakar Mantripragada: uh data set based on what you wanted to Rajashekar G: part. Satyasri Prabhakar Mantripragada: show. Um what I meant is questions Rajashekar G: Oh, Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: we are doing it with Satyasri Prabhakar Mantripragada: detction. Rajashekar G: us. Satyasri Prabhakar Mantripragada: Wait, wait. Already. Rajashekar G: Plant head questions. Which plants are underutilized versus overutilized? Satyasri Prabhakar Mantripragada: Huh. Okay. Huh. 01:08:27 Rajashekar G: Questions? Satyasri Prabhakar Mantripragada: Yeah, at least we target history. Rajashekar G: and visually like uh somethingual Satyasri Prabhakar Mantripragada: Then Rajashekar G: protocol. Satyasri Prabhakar Mantripragada: check. Sorry. data set temperature. Rajashekar G: temperature machine temperature basically. Satyasri Prabhakar Mantripragada: So question one of the questions defect spike rate correlated with stand suppose machine vibration plus temperature drift. Rajashekar G: Okay. My temperature and they are in the Satyasri Prabhakar Mantripragada: Uh so defect Rajashekar G: column. Satyasri Prabhakar Mantripragada: spike. Rajashekar G: I put this in the shape. Satyasri Prabhakar Mantripragada: Um if we have to propose a solution uh to the prospects how do we have to suggest like should it be onrem or on Yeshwanth Reddy Yerraguntla: H Satyasri Prabhakar Mantripragada: cloud detection deployment Yeshwanth Reddy Yerraguntla: deployment aspects. What's going on? Satyasri Prabhakar Mantripragada: approach. Yeshwanth Reddy Yerraguntla: Um, document. Satyasri Prabhakar Mantripragada: document some somehow you need to feed her what it takes to generate build this kind of an application diagram. You may even describe her or even make her understand verbally anything. Okay. 01:11:48 Satyasri Prabhakar Mantripragada: presentation better actually. Maybe she can directly talk to the slide. Rajashekar G: Don't know what to do. upcoming shot. So what I'm trying to do is sheet okay Google sheet I'll see how many mission ID types are Yeshwanth Reddy Yerraguntla: H.B. Rajashekar G: there uh Come Yeshwanth Reddy Yerraguntla: Mhm. Mhm. Satyasri Prabhakar Mantripragada: This is how it is suggesting to give a decision enabled output. And there So challen parameters do we need to monitor or capture so that we can give these kinds of insights. Yeshwanth Reddy Yerraguntla: They have to chain it back 1.8 cr last month because of improper shipping. There will be some set Satyasri Prabhakar Mantripragada: Due to Late Detection of RO defects Yeshwanth Reddy Yerraguntla: of Satyasri Prabhakar Mantripragada: late for example the machine roll was not maintained like was not serviced which caused the wear and tear to increase wear and tear Rajashekar G: sleep. Satyasri Prabhakar Mantripragada: defect. Yeshwanth Reddy Yerraguntla: See, we are currently defect detecting uh issues, Rajashekar G: Good job. Yeshwanth Reddy Yerraguntla: wrong detection last month due to 01:18:11 Satyasri Prabhakar Mantripragada: Mhm. Yeshwanth Reddy Yerraguntla: detction detection detect. It could not detect in process A. What we detected in process Satyasri Prabhakar Mantripragada: H Yeshwanth Reddy Yerraguntla: B or are we going too specific into this one example? Satyasri Prabhakar Mantripragada: guess what you said is correct. As for our current scenario, Maybe it could be that um say there are four stages or five stages within the production. You detected the defects towards the last stage while you could have Yeshwanth Reddy Yerraguntla: Honor. Satyasri Prabhakar Mantripragada: avoided in phase two. But as of now maybe phase two we may not have a visual based defect detection system. It is something else an X-ray or ultrasound or sound or something something. This software cannot Yeshwanth Reddy Yerraguntla: Got Satyasri Prabhakar Mantripragada: help. Yeshwanth Reddy Yerraguntla: it. Satyasri Prabhakar Mantripragada: prevention at the same time. Yeshwanth Reddy Yerraguntla: See level questions. Satyasri Prabhakar Mantripragada: Okay. Yeshwanth Reddy Yerraguntla: Yeah, I mean not exactly these questions but uh qualitative I was seeing similar set of questions that are being 01:21:07 Satyasri Prabhakar Mantripragada: Hey, Yeshwanth Reddy Yerraguntla: combination single data set I think should be fine in the contain what we saying is we went through data sets but we couldn't find uh uh diverse data in uh these data They all are talking about IoT sensit. Satyasri Prabhakar Mantripragada: Okay, one Rajashekar G: So one record Satyasri Prabhakar Mantripragada: shot. Rajashekar G: in which we are having 50 machine ids. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: 50. So it is like 93% Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: I Yeshwanth Reddy Yerraguntla: 1.3% in low 93. Satyasri Prabhakar Mantripragada: 93. Rajashekar G: 3%. So high efficiency high. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Okay. error max 10% 13% let's say 43% it's in low efficiency production speed units per Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: How about that? Rajashekar G: multiple columns depend I check Yeshwanth Reddy Yerraguntla: Yes, correct. Rajashekar G: okay straight forward column matches something Yeshwanth Reddy Yerraguntla: Basically Rajashekar G: else. Yeshwanth Reddy Yerraguntla: groupations until you try it what it's going to solve it upload solve Rajashekar G: Okay. All link. Yeshwanth Reddy Yerraguntla: One 01:23:39 Rajashekar G: Karma. Yeshwanth Reddy Yerraguntla: second. Uh, IoT data Rajashekar G: Vision defects. Yeshwanth Reddy Yerraguntla: set Rajashekar G: Vision defect data set. Awesome. Yeshwanth Reddy Yerraguntla: CSV. I will upload it link. Rajashekar G: Okay, I'm sharing. Yeshwanth Reddy Yerraguntla: Yeah. I search Rajashekar G: file shutters. Yeshwanth Reddy Yerraguntla: you. Rajashekar G: Try that. Yeshwanth Reddy Yerraguntla: Okay. Same data set. Rajashekar G: Same one. Yeshwanth Reddy Yerraguntla: I know 6G in there. Rajashekar G: The name 6G network data including latency and packet loss. Packet loss in. What is this? This is it. Excel Yeshwanth Reddy Yerraguntla: upload. Yes. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: screen. Rajashekar G: Okay. I stopped. Yeshwanth Reddy Yerraguntla: What's up Rajashekar G: So what it is saying like packet packet loss and uh network latency that may also directly or indirectly Because this inefficiency Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: server speed it is getting slower. If packet loss is more so few of the chunks are falling in between like not even reaching the server. 01:27:58 Rajashekar G: The information Yeshwanth Reddy Yerraguntla: Okay, got Rajashekar G: effect. Yeshwanth Reddy Yerraguntla: Okay. Sample question on the you can use this going Rajashekar G: Thank you. Yeshwanth Reddy Yerraguntla: forward. Rajashekar G: defects by correlated. I'm sending the question Yeshwanth Reddy Yerraguntla: with me. Rajashekar G: stand three and vibration and temperature difference. Yeshwanth Reddy Yerraguntla: Check. Rajashekar G: solve it. They're not making it public. Okay. Yeshwanth Reddy Yerraguntla: Go ahead. Uhhuh. quality control effect rate with itself obviously 100% there's no correlation High level. Rajashekar G: Very questions. Sorry directly like let's say in the month of Jan um which machine causes me more error or the Yeshwanth Reddy Yerraguntla: I'm waiting for some Rajashekar G: defects Yeshwanth Reddy Yerraguntla: question. So this ticket yesterday I had a discussion with and today I'll start creating tickets for the updating the document. Please. Thank you. See it was able to it is able to answer now based on pure 01:32:30 Rajashekar G: You're Yeshwanth Reddy Yerraguntla: analysis. Rajashekar G: right. Yeshwanth Reddy Yerraguntla: Maybe Rajashekar G: like which machines are causing which machines were under maintenance for more time or which machines I need to replace to get my productivity increase. Yeshwanth Reddy Yerraguntla: I think I think so. Rajashekar G: I'm one month. Yeshwanth Reddy Yerraguntla: So this is a I think HDMX issue react. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Good thinking. Rajashekar G: HTMX server Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: side. Yeshwanth Reddy Yerraguntla: That's So answers is not the issue. Rajashekar G: No. Yeshwanth Reddy Yerraguntla: We how do we set up the infra to get the answers? Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Are you there? Right. Rajashekar G: So solve it is doing based on Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: E tools are Yeshwanth Reddy Yerraguntla: Run Python code. Pandas enabled Python code but Rajashekar G: okay. Yeshwanth Reddy Yerraguntla: instructions. Rajashekar G: CSV file Yeshwanth Reddy Yerraguntla: Uh Rajashekar G: JS Yeshwanth Reddy Yerraguntla: instructions answer Rajashekar G: character. Yeshwanth Reddy Yerraguntla: whatever pandas whatever you want to call it success rate But uh 1 or two I will set up a small uh this thing uh yeah 01:35:51 Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: 9:00 this is what we are able to answer these are the types of questions Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: feedback. Rajashekar G: Okay. I Yeshwanth Reddy Yerraguntla: See Rajashekar G: mean, Yeshwanth Reddy Yerraguntla: roots.com Rajashekar G: oh no. Yeshwanth Reddy Yerraguntla: you can create your own things Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: folder file out there, Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: right? So, Rajashekar G: fine. Yeshwanth Reddy Yerraguntla: we can always create fresh things. Yeah. The Rajashekar G: Got it. Yeshwanth Reddy Yerraguntla: public Rajashekar G: Okay. Morning 7:30. Discussion meeting architecture. Yeshwanth Reddy Yerraguntla: Easy. Satyasri Prabhakar Mantripragada: So far department Yeshwanth Reddy Yerraguntla: Uh just trying to understand department. Sorry. Satyasri Prabhakar Mantripragada: just to keep you informed. latest. Yeshwanth Reddy Yerraguntla: What I will do? Satyasri Prabhakar Mantripragada: We can leverage his time Yeshwanth Reddy Yerraguntla: Got it. Satyasri Prabhakar Mantripragada: schedule Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: better. comments. Okay. Yeshwanth Reddy Yerraguntla: connect 01:38:26 Rajashekar G: Oh, Yeshwanth Reddy Yerraguntla: with. Rajashekar G: add another Satyasri Prabhakar Mantripragada: Huh? Yeshwanth Reddy Yerraguntla: Got it. based on my confidence. Satyasri Prabhakar Mantripragada: 9:30. Rajashekar G: Okay. Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: And continue working for example. Satyasri Prabhakar Mantripragada: Okay. Not related to this. Um I wanted to understand prompt for applications to Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: test and rag Yeshwanth Reddy Yerraguntla: prompt for first time. Rajashekar G: Copy. Satyasri Prabhakar Mantripragada: whatever we are building we need to test them Yeshwanth Reddy Yerraguntla: Yeah. All Satyasri Prabhakar Mantripragada: unstructured. Yeshwanth Reddy Yerraguntla: right. Satyasri Prabhakar Mantripragada: So there is a simple npm or something which where you can define your own parameterized uh prompts and give expected outcome. Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: Start exploring prompt Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: four. Yeshwanth Reddy Yerraguntla: Yeah, I Rajashekar G: Autoate. Yeshwanth Reddy Yerraguntla: just sorry. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I think that we want to discuss yeah WhatsApp that I will be ready with so and so questions 01:41:02 Satyasri Prabhakar Mantripragada: So basically I wanted to understand if you are building any LLM then we should test them Yeshwanth Reddy Yerraguntla: and Satyasri Prabhakar Mantripragada: properly just like software. If it is third party library we don't have to test but if anything LLM or Yeshwanth Reddy Yerraguntla: he Satyasri Prabhakar Mantripragada: that we are building then we we need to find out tools to test them. Yeshwanth Reddy Yerraguntla: 100%. Satyasri Prabhakar Mantripragada: Uh so at least I'll also find out what tools can we Yeshwanth Reddy Yerraguntla: Yes. Satyasri Prabhakar Mantripragada: use and do hands on and try to figure it out on how to Yeshwanth Reddy Yerraguntla: got Satyasri Prabhakar Mantripragada: test. I need to upgrade and find out new ways of testing Yeshwanth Reddy Yerraguntla: anything else. I think we Satyasri Prabhakar Mantripragada: things. Yeshwanth Reddy Yerraguntla: can but for sure we need these kinds of tools and for sure Satyasri Prabhakar Mantripragada: just going through um Yeshwanth Reddy Yerraguntla: Dspot. Satyasri Prabhakar Mantripragada: okay Yeshwanth Reddy Yerraguntla: You can let us know Satyasri Prabhakar Mantripragada: DSP py Yeshwanth Reddy Yerraguntla: DSP. Satyasri Prabhakar Mantripragada: Okay. 01:42:18 Rajashekar G: Make Satyasri Prabhakar Mantripragada: Okay. Yeshwanth Reddy Yerraguntla: Round four. Satyasri Prabhakar Mantripragada: I shall try it Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: rag modeling. I heard about it but I didn't Yeshwanth Reddy Yerraguntla: I guess I guess Yeah. Satyasri Prabhakar Mantripragada: uh if we are building it then I can test I don't know how to test them or if we have any LLM also I should be able Yeshwanth Reddy Yerraguntla: We are building rag. Yeah. Yeah. Satyasri Prabhakar Mantripragada: to Yeshwanth Reddy Yerraguntla: That how to test it. We'll try to expose it. Very important. Satyasri Prabhakar Mantripragada: inco Yeah. If you know any openl is there if you can find out give us a small scale LLM so that we can just experiment. Yeshwanth Reddy Yerraguntla: H. Satyasri Prabhakar Mantripragada: Basically any output should be predictive deterministic. So you give this that okay you certify with that then you'll understand okay you are able to you understood how to use the tool not that you have tested the LLM Yeshwanth Reddy Yerraguntla: Mhm. 01:43:56 Yeshwanth Reddy Yerraguntla: Got it. Satyasri Prabhakar Mantripragada: Uh-huh. Okay. Yeshwanth Reddy Yerraguntla: And we have one Satyasri Prabhakar Mantripragada: Okay. Yeshwanth Reddy Yerraguntla: GPU just that we are not actively using Satyasri Prabhakar Mantripragada: Mhm. Yeshwanth Reddy Yerraguntla: it. Satyasri Prabhakar Mantripragada: desktop machine. Rajashekar G: I completely Yeshwanth Reddy Yerraguntla: What? Satyasri Prabhakar Mantripragada: I I see you beep sound. Rajashekar G: Oh, Yeshwanth Reddy Yerraguntla: hardware. Yeah. Rajashekar G: heat. Satyasri Prabhakar Mantripragada: me if you get a chance to talk to Naven just tell him from your side also. I did not put it to him but I wanted to talk. Uh but you are more authorized on this is that uh the AI team is complaining that they their machines are too slow. At least brought it to my notice that it is too slow and Yeshwanth Reddy Yerraguntla: H. Satyasri Prabhakar Mantripragada: uh better mach Rajashekar G: One Yeshwanth Reddy Yerraguntla: on. Rajashekar G: sec. Satyasri Prabhakar Mantripragada: I'll also talk to but just put him a word that okay this is what it Yeshwanth Reddy Yerraguntla: Got it. 01:45:40 Yeshwanth Reddy Yerraguntla: Sure. Satyasri Prabhakar Mantripragada: Looks like Yeshwanth Reddy Yerraguntla: Got it. Sure. Satyasri Prabhakar Mantripragada: Yeah. Yeshwanth Reddy Yerraguntla: Sure. Satyasri Prabhakar Mantripragada: Not Rajashekar G: some old laptops. Yeshwanth Reddy Yerraguntla: Oh no. Okay. Oh. Soap. Satyasri Prabhakar Mantripragada: employees Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: rotate. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: previous. Yeshwanth Reddy Yerraguntla: Okay. And now sir, Satyasri Prabhakar Mantripragada: Okay. So, Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Salesforce Yeshwanth Reddy Yerraguntla: s Satyasri Prabhakar Mantripragada: questions. Rajashekar G: manual. Yes. Satyasri Prabhakar Mantripragada: Do we know Rajashekar G: Stop. Satyasri Prabhakar Mantripragada: Okay. Fine. Rajashekar G: Starting. Satyasri Prabhakar Mantripragada: Okay. Okay. Fine. Sorry. Any additional questions? So final Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I'm Rajashekar G: And sir like uh Satyasri Prabhakar Mantripragada: chest Yeshwanth Reddy Yerraguntla: ch Rajashekar G: demo. It should be having two sources like Salesforce and insurance DB or only Salce is fine. Satyasri Prabhakar Mantripragada: no Salesforce insurance DV Gmail does not make sense to an external 01:47:56 Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: user. Rajashekar G: Right. Okay. Satyasri Prabhakar Mantripragada: Visual Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: effectual Rajashekar G: looking. Yeah, got it. Satyasri Prabhakar Mantripragada: detect Yeshwanth Reddy Yerraguntla: lineage. Uh Satyasri Prabhakar Mantripragada: defects. If somebody asks with a prompt that what caused these defects and it should go backtrack to a stage where this is where the root cause is. Yeshwanth Reddy Yerraguntla: fake data already. Demo Satyasri Prabhakar Mantripragada: Just Rajashekar G: Madam Yeshwanth Reddy Yerraguntla: already. Rajashekar G: in our current demo I'm sharing the screen Satyasri Prabhakar Mantripragada: better presentation. Hey Please vision detect. Vision defect. Rajashekar G: Uh, Satyasri Prabhakar Mantripragada: Visual defect. Rajashekar G: I'll see. Okay. Satyasri Prabhakar Mantripragada: Okay. Yeshwanth Reddy Yerraguntla: Neat. Rajashekar G: Not found. So like this is a root cause. Satyasri Prabhakar Mantripragada: Mhm. Rajashekar G: Poor Satyasri Prabhakar Mantripragada: Okay. Rajashekar G: filtration like the issue can be material on the top. First machine related Satyasri Prabhakar Mantripragada: H Yeshwanth Reddy Yerraguntla: What Satyasri Prabhakar Mantripragada: text visual representation with drill down Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: capabil when I hover on a certain stage then it'll give you its own information. I want to expand. Rajashekar G: design Satyasri Prabhakar Mantripragada: Uh Yeshwanth Reddy Yerraguntla: do you see? Rajashekar G: and they wanted to hersa wanted to design system complete Satyasri Prabhakar Mantripragada: okay. Rajashekar G: experience stopped working this. Satyasri Prabhakar Mantripragada: Okay. Fine. Rajashekar G: Yeah. Satyasri Prabhakar Mantripragada: Global root cause analysis Rajashekar G: Ei Satyasri Prabhakar Mantripragada: changes. Rajashekar G: Answer. Satyasri Prabhakar Mantripragada: Hello. Let me voice in Yeshwanth Reddy Yerraguntla: There. Satyasri Prabhakar Mantripragada: Okay, fine. Then uh we'll end this. Yeshwanth Reddy Yerraguntla: Yeah, Satyasri Prabhakar Mantripragada: H Yeah. Yeshwanth Reddy Yerraguntla: man. I check it. Huh? Transcription ended after 01:53:44 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.