GraphStuff.FM: The Neo4j Graph Database Developer Podcast

Providing Better Business Intelligence with Vish Puttagunta

Episode Summary

Joining us is Vish Puttagunta, the CEO of Power Central where they are bringing ERP, data & intelligence together in the food industry. He is a Senior Analytics leader with proven track record of applying Data, Artificial and Business Intelligence techniques to generate tangible and measurable ROI in Business, with a firm focus on Food Manufacturing and Food Packaging.

Episode Notes

Tools of the Month:




Episode Transcription

Alison Cossette: All right. We are live and we will get going. Welcome back, graph enthusiasts, to GraphStuff.FM, a podcast about graphs and graph-related technologies. I am your host, Alison Cossette, and I'm joined today by ABK of Neo4j and Vish, the CEO of Power Central. Joining us, Vish today. He is the CEO of Power Central where they're bringing ERP data and intelligence together in the food industry. Vish is a senior analytics leader with a proven track record of applying data, artificial and business intelligence techniques to generate tangible and measurable ROI in business with a firm focus on food manufacturing and food packaging. Vish, welcome. We're so happy to have you today here at GraphStuff.

Vish Puttagunta: Thank you for having me, Alison.

Alison Cossette: We met recently at Data Day Texas and we had some really interesting conversations. I was wondering, can you tell us a little bit about the origin story of Power Central?

Vish Puttagunta: Yeah, sure. I worked at TI, Texas Instruments, Samsung as a semiconductor software engineer, so I used to do a lot of signal processing with audio and speech. About 2014 or so, I transitioned into analytics and data science for the enterprise. About 2015, 2016, that's when I started getting introduced to Neo4j along with Elasticsearch and the other big data technologies. I've had my mind on Neo4j for a very long time and to build a killer app since then. One of the things that really frustrated me with data science as a whole was that the underlying data usually comes from an ERP or a CRM or some kind of a transactional system underneath, and the data was often very unreliable and we were just making stuff up, right?

I was just asking like, "Why is that? Why is the underlying data so terrible?" What actually ended up happening was that when I interviewed a lot of SAP guys that were implementing the ERPs, they were like, "Hey, man. The cost of implementing any of these customizations for the actual warehousing guys..." For example, if you look at an ERP, I would say in a food manufacturing company about 60 to 80% of the data actually comes from the warehouse receiving guy, the production guy, and the shipping guy, and they were not being helped correctly. The user interface for those guys was not very easy, and that's the reason for all the data gaps in the system. That's what I observed, and so in 2019, 2020, I decided to go vertical, so instead of just being in data analytics and AI I said, "You know what? Let me actually build a complete vertical."

I ended up buying a food packaging company here in Dallas. One of my buddies, they owned a food packaging company and they were going gangbusters, they were growing 20% month over month, but they were really struggling. I mean, one of the biggest problems they had was they were supplying to the meal kit industry, like HelloFresh and Blue Apron, and they had almost 1500 SKUs by the time they even got to half a million dollars in sales. I mean, to give you some perspective some of the larger clients we have have maybe 30, 40, 100, 200 SKUs and even then they will struggle doing the procurement planning and production planning. These guys had like 1500 SKUs that they had planning issues with.

It was really nice because for the first time, I could put away the BS. When we go into data science and ask consulting questions, people don't really give what really matters in their profit and loss statement or their balance sheet or cash flows, but when you own a company... I mean, I'm a minority owner in the company. In the end, I have access to the financials so there's no BS. I know, "Hey, this is what's going on. Here's the growth in our sales and here's the impact on our balance sheet." I was able to apply technology where it mattered most as opposed to guessing and really putting months and months of effort and only realizing that, "Oh, it's a cool problem, it's a nice problem, but it does not impact our bottom line or our top line."

Alison Cossette: Yeah. Fascinating. Tell me a little bit about once you got in. You came in knowing that you had some type of a graph problem, so can you tell me a little bit more about once you got in, what did you find within graph itself and was the data modeling easier because you were in... What was that process like?

Vish Puttagunta: Yeah. I mean, I did not make such assumptions. I had a lot of tools in my arsenal from traditional statistical forecasting modeling to deep learning to graph-based algorithms. I had all these things in my head, so I came with an open mind to this company saying that, "Okay, how can we first..." My first objective was like, "Okay, let's keep the data really sane and let's make it really easy for the data to come in and then solve problem one by one." Actually, the first problem that I solved was the cash flow problem. This company was doing about half a million dollars in sales and losing anywhere between 30 to $50,000 every month, so they were not able to just establish their cash flows.

The first thing I had to do was to really just very simple business intelligence to understand what is their cost of goods sold. A lot of times, the problem is it sounds like a very simple, obvious answer, but unless you have your ERP set up correctly and people are using it correctly, you will not know what your cost of goods are. That was the first thing that was a game changer. Previously, imagine the plight of a CEO that is actually helping grow the company 20% month over month and his investors not trusting him because his books are not in order, he's going every Thursday and basically asking for $200,000 and it really spooked the investors. That was the first problem I had to solve.

Then the second... Once I had this ERP and mobile apps and business intelligence reports, then the investors, the financial head of the company, the people who are giving us loans, they all had access to really good reporting that instantaneously gave them what the margins were. You wouldn't believe most of them... Our CEO had a general idea about how much money he's making on the top three products and how much money he was losing on the bottom three products, but there were a thousand products. So our average gross margins were terrible, 25% or something like that, and it was very inconsistent. On one product we were making 35 to 40%, which was our target, and on some products we were losing 45% because freight would fluctuate or raw material would fluctuate during COVID. That was the first problem I had to solve, yeah?

Alison Cossette: Yeah.

Vish Puttagunta: Then I had to solve a problem of as the CEO was delegating more and more to other folks, the other folks in procurement, you could make some serious mistakes. If you go and basically buy a product at a price that is not affordable, you will end up losing money on the job. One of the primary... What led me to Neo4j, I was just waiting for... Hey, I knew that Neo4j is really cool. I wanted to always apply that technology, but what really lended itself, think about any food manufacturing or food packaging company. You have sugar coming in, you have 10,000 lbs of sugar that's coming in three days from now, and suddenly during COVID and even now, the procurement... If the vendor calls in and says, "Hey, man. I'm not going to make that 10,000 lbs delivery three days from now. I think it's going to take another three weeks for me to deliver that."

Now what? Because the sugar could be going into a syrup, that syrup could be going into a protein bar, the protein bar could be going into a case pack, and all of these are different jobs and on different days so there's a ripple effect. There's a network effect when that happens. The traditional MRP production management and material requisition planning tools that are available in the ERP, they were not interactive, number one. They were not showing the color codes. For example, I could not show a single screen to my procurement manager, to my production manager, and my sales manager on a single screen and just highlight, "Hey, what are the risks in the next three months because of the sugar coming late?"

Alison Cossette: Right.

Vish Puttagunta: A lot of times, these guys have a lot of knowledge, they have a lot of domain knowledge, they've built a lot of relationships with their vendors and with their customers, so they can be very nimble if you highlight the risk to them because sometimes the right thing to do would be probably call the customer and say, "Hey, man. I can't deliver this a week from now. Can you take this delivery two weeks from now?" But sometimes that's not an option, so sometimes you basically have to say, "You know what? Instead of buying sugar from this vendor at dollar per pound, I'm going to buy it from this vendor at $1.15 per pound. I know I'm going to take a hit on my gross margins, but I can't slip this order up."

Alison Cossette: Got it.

Vish Puttagunta: These are the kind of things that we really had to visualize, and as you can see they are all different items, different production orders, and there's a ripple effect there. When I looked at it, it just screams graph at me. Whenever there's a recursive join, that screams graph at me, so I was like, "You know what? I think this is a great situation for graphs, so let me build a graph tool and just show them this risk." As a matter of fact, none of my customers even know what a graph database is. They don't even know. For them, it just looks like a timeline and a risk and that's what they're used to.

Alison Cossette: Interesting. How do you visualize that? Do you... From a graph perspective, do you just show the path over time of how those fall through or what does that look like?

Vish Puttagunta: Yeah. Let me actually show that to you over here. Let me... How do I...

Alison Cossette: Down at the bottom of your screen in the middle, you should see a share screen button.

Vish Puttagunta: Okay. Yeah. See share screen button. Entire screen. System audio. I'm not getting... The share is disabled.

Andreas Kollegger: Oh, really? Okay.

Vish Puttagunta: Oh, can you [inaudible 00:10:57]?

Andreas Kollegger: Yeah, I'm trying to look at that. It should be.

Vish Puttagunta: Okay, got it.

Andreas Kollegger: There you go.

Vish Puttagunta: Okay. Can you see my screen?

Andreas Kollegger: I think I now need to just...

Vish Puttagunta: Yeah. there we go.

Andreas Kollegger: There we go. There we go.

Vish Puttagunta: Yeah. Basically, this is how typically our interface actually looks like. If you look at our orders, so here we have orders going in. For example, I know that there's a... I got to deliver a thousand of this particular item, I got to deliver... These are the orders that are coming in for different items, and these are the production orders in my system and these are all the purchase orders in my system and these are all the stock levels in my system. For example, the case that I just mentioned to you, let me actually narrow this down to a single machine so that I don't have all the clutter.

There are different ways of actually narrowing down what I'm actually seeing over here because typically our customers, they have 10 or 15 production lines and they're planning what can be done on a specific machine is pretty unique, so usually about five or six SKUs can be done on a single machine. They go machine by machine and say, "Okay, let me plan out production for this machine for the next three months or something like that." When they look at a specific machine, they know that this item can be made on that machine, this is what the demand is, and everything is color coded saying that, "Okay, if I produce everything in time, then these things are going to be able to deliver in time." But say, for example, here are my purchase orders and say, for example, the vendor says, "I can only deliver 800 lbs. I cannot deliver 8000 lbs." You see how it's actually automatically updating the stock levels and updating the assembly orders over here?

Alison Cossette: That's fantastic.

Vish Puttagunta: That's how people know that, "Oh, there's going to be a problem with this production order." So now the production manager is like, "Okay, what is the impact of this? What is the impact on the sales order if I move this assembly order to the eighth of the month?" I can play a lot of what-if scenarios, and, as you can see, the sales orders are now highlighted in red.

Alison Cossette: Fantastic.

Vish Puttagunta: It just looks like a very simple timeline over here that shows what is really going on.

Andreas Kollegger: But that's so cool. Think about what you were describing earlier. It feels like this is simply moving from having no sight to just having some sight. It's not even to the point of insight yet. It's like, "Okay, you didn't even know what was happening in the business, but now you can see it. Now you can actually understand it."

Vish Puttagunta: Now you can actually see it, and also this thing is so seamless because when I did this movement from assembly order from here to here, it goes and does it right back into the ERP. It is seamlessly integrated into the ERP because this orders, the purchase orders, the inventory levels, the production orders, all of them are basically real deals, real sales orders in the ERP, so it's seamlessly connected to the ERP.

Alison Cossette: That's fantastic. It's always interesting to me how folks leverage the graph for business users because oftentimes from underneath, you... We being in graph, our tendency is to like, "Oh, let's show you the graph." But so much is possible without the users even knowing that it's graph, so using it as that tool underneath, it doesn't have to look like circles and lines. We can present it in ways that people are used to seeing things.

Vish Puttagunta: No, no. I mean, what was really interesting is we gave them a lot of... Initially, you think about edges and nodes and that's how you want to visualize and show it to the customer. Then for the first almost six months, we were trying to show them how the graph works and here are the nodes and edges, and my customers were very confused. They were too overwhelmed like, "This is not how I think." I asked them like, "How do you plan? You are doing your planning today. You don't have a graph database or anything like that. How do you do your planning today?" Most of them came up with an Excel sheet that actually looks like what I actually showed you.

Alison Cossette: Yeah. Really?

Vish Puttagunta: The only problem was that Excel sheet had to be updated almost every day. They had to go to the ERP, look up the data, and then put the data into the Excel sheet. It was just a lot of hours of updating that Excel sheet, and then they had Excel formulas to do the color coding and everything, but if you think about a recursive model, like the sugar going into the syrup into a bar, they had to literally make those calculations and put those calculations and update those inventories manually and then they could see what they could see. What they were doing and what they're able to do now, so now they just have a 15-minute sprint meeting every day. They're meeting every day. That's all. The sales guy, the purchasing guy, and the production guy, they would just together have a 15-minute meeting, and usually the planning takes anywhere from 30 minutes to an hour every week. That's it. I mean, they're able to plan three months out for production order in less than about 30 minutes to 40 minutes.

Alison Cossette: That's phenomenal. Phenomenal. Go ahead, ABK.

Andreas Kollegger: What I love about this entire story... For me, I know very little about the food industry, of course, and it feels like so many of the verticals that you might be able to get into... If you haven't thought about it before, you're like, "Well, I don't even know what happens there." But then you describe things like that parts of the business running on a spreadsheet, like, "Okay, actually I've heard that before." In every industry there's an embarrassing amount of the process that's usually centered on a spreadsheet somewhere, but then this whole kind of process that everyone is still going through of taking businesses that you know there's actually... Once you start to pull a thread and you look at that spreadsheet and what's happening, you take it one step further, the thread leads you ultimately to the entire business, and you're like, "Okay, as you've done, if we can just see all of that at once rather than the pieces, now you can start to reason about it. Now you can actually do things with it."

It feels like... Even as you were describing it, okay, I went from knowing... At the beginning of this session, I knew nothing about the food industry. I don't know that I could give a talk on it now, but now I'm like, "Oh yeah. That makes sense. I see where this would lead to this next thing." The entire supply chain, everything you're describing about, makes such intuitive sense, but you have to think about it first. I can imagine that... Now I'm going to back up a little bit and say have a bit of empathy, I suppose, for folks on the ground who are just getting things done. They don't have to take the time necessarily. They don't realize how much time they're spending on things that they don't need to be doing or the impact of things that happened because they don't have this vision and like, "Well, that's data stuff. Maybe we won't get to it." But now that you have a tool like this, you can bring to them or think anybody in the industry who sees this, the light bulb was going to go off and be like, "Oh my God, we want that. Could we have that?" Has that been your experience as well?

Vish Puttagunta: There are two things that I really... I keep telling people I started this company because I had too many fights with the ERP people. The problem that I always saw was that the guys who are ERP, I'll give them credit for they understand process. Anybody that is really good in ERP and making those 400 bucks an hour, they really understand process and they can really talk well and really fish out the problems from the customers. The problem often tends to be they don't understand anything beyond that ERP. They don't understand the difference between when to use the ERP tool because ERP is your single source of truth if you implement it correctly. That's about it. You don't want it to be super intelligent because you're limited by the language of what that ERP is, whether it's an app in SAP or it's AL code in Business Central. You're limited by that language, right?

Andreas Kollegger: Mm-hmm. Mm-hmm.

Vish Puttagunta: But when you have to solve certain problems, you want to take a step back and say, "Hey, is this the right tool for this job?" You will often find that, no, if you just want very quick metrics, Power BI is a hundred times better than AL or if you want iterative tools or these kind of supply chain, understanding these bottlenecks and stuff like that, graph is a beautiful way of doing it. Okay, so let's use graph to do it. That's what's really missing, and the other problem is that, to be honest, when I worked as a data scientist, I did not understand the ERP. I did not understand which table has what data, who's putting what data in this particular thing in this particular table. It's a big problem in the industry. Then these big data analysts and engineers, they're paid $200 an hour. I was paid $250 an hour.

The biggest challenge is when you're getting paid $250 an hour, you better be a magician. But, again, I'm just a normal person that happens to know these mathematical tools and these graph tools, but I don't know how, for example, Chili's makes money or this makes money, and who put the data in the ERP. There's a huge learning curve there. I literally had... I was so frustrated, I had to go bloody buy a packaging company and get on their board, sit on their warehouse floor for a whole year to understand this. I keep telling people... When people corner me, "Are you Power..." I had a failed consulting company. I just couldn't scale it. It was not fail per se, but I couldn't scale it beyond myself. I would work for 100 hours, I would get paid for 100 hours. I could not scale it. The biggest problem was a lot of the customers, they want to solve a problem. Nobody cares whether you use Neo4j or whatnot. In order to solve the problem, you got... The third problem is that you got to know what it's worth for them because there's a lot of cool stuff you can do with Neo4j, but you got to understand how much people are willing to pay for it, right?

Andreas Kollegger: Mm-hmm.

Alison Cossette: Right. I mean, ultimately it has to have some impact on the process of the business that's going to help you make more money or be more efficient, right?

Vish Puttagunta: Right.

Alison Cossette: It has to have a tangible impact.

Vish Puttagunta: Yeah. For example, for our customer I used to spend a lot of money on cool stuff in my previous company. If somebody says, "That's a cool idea." I would go and spend 15, $20,000 building a prototype and stuff like that, only to realize when I went back to them and showed them the tool and they would say, "Oh, are you willing to pay $1000 per month for it?" They'd be like, "Man, it's not worth that much." I'm like, "Why did you tell me that was a problem?" How many of these proof of concepts have been going on in the enterprise where it is cool, but if you ask them like, "Show me how this impacted on your P&L," You never get a straight answer?

Alison Cossette: Yeah. Yeah.

Vish Puttagunta: So, I went to my customer this time and I was waiting for them to hit that breakneck point because one of my customers, they went from 15, $20 million to... They were pushing that $40 million boundary, and now they were planning... Normally they would be using these Excel sheets to plan about two, three weeks out, but now they have to plan three months out. They have to see that there's a supply chain issue that's going to happen three weeks out like, "Hey, I'm..." Or they were ordering so many things, but sometimes the vendors would not acknowledge and confirm the orders and they wouldn't realize that, "Oh, I sent this order a month ago and now it's two days before production and they haven't even received that order. They haven't even acknowledged that order." Then the production manager came back and said like, "Hey." She does procurement, production, everything, and she's like, "Vish, I'm getting overwhelmed. At this point of time, we are planning on hiring a planner that would cost us $90,000. Either that or you solve the problem for us." Okay, now I know that it's a $90,000 problem, right?

Andreas Kollegger: Mm-hmm.

Alison Cossette: Got it.

Vish Puttagunta: That is how I know that if they hire that person, it's a $90,000 expense on their overhead GL, on the salaries general ledger. Now I can say, "Hey, can you pay me $12,000 a year? I will solve this problem for you. Thousand bucks a month. Does this sound fair?"

Alison Cossette: Yeah.

Vish Puttagunta: "Yeah, it sounds fair. Let's go do it."

Alison Cossette: Oh, I love it. I love it. What I really appreciated about Vish, why I really wanted to have him on here, is I really appreciate the way that you envision the problem and are so tied to that tangible outcome and it's all very... I don't know what the appropriate word is, but it's so real. Everything that you're doing is about the actual impact, the decisions that are made, the way things move. I also come from data science, so I have empathy for massive different tables and not even being able to know where to start and help, but I think there's so much to be said about being really focused on impact when you're coming in and evaluating tools. I mean, as data scientists we hear all the time, like you said, about the POCs and what is the actual ROI on something? Why don't things get into production? I just... There's just something to me that's really valuable about your approach and the way you see it.

Vish Puttagunta: Yeah. You know, I've been really inspired by Warren Buffett. One of the things that... I watched too many Warren Buffett videos on Google and YouTube. One of the things that really struck me was he doesn't know how to predict the financial markets. He never worries about it. He always tries to buy a company that has somehow survived a catastrophe, like a supply chain disaster or something like that, so he knows that a company has to be resilient. It's a well-managed company is a resilient company, so there's always a balance. As a data scientist, I want to predict the future. Sometimes I can do that, but you got to make the people resilient, and resilience comes from highlighting the risk to the right people and making that decision actionable.

That's what we saw. How many AI models failed during the pandemic because the data that we were using was just biased towards when things were going normal and all of a sudden things are failing left and right, their forecasting was completely BS? You can't rely on that. If you look at the traditional MRP tools, they are so... You got to put all these bloody thousands of rules inside the ERP, "Oh, who are the vendors for this particular item? Oh, what is their minimum order MOQ values? Hey, what are the pricing?" They have a million gazillion rules in that MRP that they try to feed into the ERP. What happens when a black swan event like COVID happens? First of all, it takes insane amount of time. Nobody has that time to put all that data into the ERP, first of all. At least my guys don't. But if I show them this visual, it's in their head and they will be able to know, "Okay, instead of going to this vendor, let me go to this vendor." So it's more resilient, it's more reactionary. You just show the best intelligence you got to the right people and let them make the decisions.

Andreas Kollegger: Mm-hmm.

Alison Cossette: That's a beautiful takeaway. I love that. That sounds like a good soundbite to me. Vish, thank you so much for all of your time. I was really excited to have you on as a guest so I'm thrilled that you made time for us. Thank you so much. Before you go, I'm just curious, can you tell us a little bit about the artwork behind you? Because I believe that I'm seeing a turtle all the way on the end. Is that a turtle?

Vish Puttagunta: Okay, so this one has a story to it. The title... I work in supply chain, right?

Alison Cossette: Yeah.

Vish Puttagunta: Oh, okay.

Alison Cossette: Sorry. We can rearrange the furniture in your house today on GraphStuff.

Vish Puttagunta: Yeah. This is actually... I bought this in Santa Fe and this is like a Native American art. You know what the title of this artwork is?

Alison Cossette: Supply Chain?

Vish Puttagunta: No, it's called A Day in the Life Of.

Andreas Kollegger: Oh, A Day in the Life Of.

Vish Puttagunta: Okay. The very first engagement I do with any of my clients is the day in the life of. I kind of go and actually interview them and ask them how their receiving works, how their production works, how their shipping works and stuff like that. That's what I bought it for.

Alison Cossette: With Power Central, you flatten that hill and you speed up that turtle. There you go. Wonderful. Well, again, thank you so much for being with us. We appreciate the generosity of your time and, more importantly, your insight and your unique view of how you can actually help an industry move forward. I'll be a little bit, it's the first time I've met someone who said, "I wanted to know how it works, so I went and bought a company." I think that's... I love that. At this point, what we usually-

Vish Puttagunta: Yeah. You don't want to ask me about ChatGPT and stuff?

Alison Cossette: That's okay. You can come back and ask me. I can help you with that one.

Vish Puttagunta: No, I mean we are actually working on ChatGPT and we've found out... I mean, you want to spend a couple of minutes talking about how we're using ChatGPT?

Alison Cossette: Yeah, I would love to.

Andreas Kollegger: Sure. Love to hear it. Yeah.

Vish Puttagunta: I don't have demos yet, but essentially when we looked at ChatGPT and the generative AI in general, what we saw that it was really good at, it's really good at natural language. Sometimes you have to tell it... One of the challenges that we see with ChatGPT is it hallucinates a lot, so we are working quite a bit on how do we reduce that hallucination. I think the secret lies in the... I was very excited that Neo4j added the vector database support to it, so I think that is really the key. We're essentially making sure that we were able to vectorize it and we're looking at closed nodes and we're... If you have a PDF document of all these documents, we vectorize that and we have all these paragraphs and nodes, but then we know that by doing the vector correlation only answer the questions from these five nodes. Don't go to the general internet to answer questions, but answer the questions from the context that you know. That is going to be very important for us going forward. There's some exciting things coming out.

From the analytics point of view, I think people are getting a little carried away because what we've observed is that the data has to be super curated and you have to build a knowledge graph. For example, a table will have a field call number. Now, is that a warehouse shipment number or is it a sales order number or a purchase order number? Nobody knows. I mean, a human wouldn't understand. You wouldn't expect a ChatGPT would understand. It doesn't understand either. So you got to have a little bit of knowledge graph with some data definitions and you have to have super curated data, and then it's able to answer questions like, "Hey, what are the sales orders at risk?" It's able to answer that, but we're building that knowledge graph as we speak. [inaudible 00:30:05].

Andreas Kollegger: That's really interesting. Can I ask, Vish, so for the PDFs you're describing, are these PDFs that are... What's the content of the PDFs, I suppose? Is it mostly text or is it tables as well or...

Vish Puttagunta: Yeah. I mean, the first thing that I asked myself was like, "What is a medium to low hanging fruit? What is not a theoretical science problem but something that we can actually work with?" It's text data. I went and asked the CEO of my packaging company, "Hey, where do you have the most amount of text in the company?" He's like, "Compliance. Compliance, standard operating procedures, and stuff like that." You're like, "Okay, do your people have trouble getting the answers they need when they have a problem?"

Andreas Kollegger: Mm-hmm.

Vish Puttagunta: The answer is yeah. For a small company they're okay, but as the companies grow bigger all this data is all over the place, sometimes they have people that speak Mandarin, sometimes people that speak Spanish and they want to get the answers in Spanish. ChatGPT is amazing at that. That's where I foresee... Especially I think the analytics will take a little time because, again, it's not what you can show to be cool, but people are making very serious financial decisions and it better be accurate. I still have a little bit of challenge trusting ChatGPT with the analytics, but on text I think we've got some really cool techniques up our sleeve to make sure that it doesn't hallucinate and it answers more of a reading comprehension than being more creative. I don't want it to be... In the supply chain, I don't want it to be creative. I want it to be accurate, you know?

Alison Cossette: Yeah.

Andreas Kollegger: Yeah. Yeah. That makes sense.

Alison Cossette: I know a lot of people are having a challenge trying to figure out with generative AI what are some of the applications where there's the big ROI? When you're thinking about translation and, like you were saying earlier, what is the value of that and does it make sense to do it, what are you bringing in as far as how you're calculating whether putting this into production is going to be valuable? How do you quantify?

Vish Puttagunta: I mean, the good news about food manufacturing, at least the sub-150 million... That's one of the reasons why we kind of focus on that 5 to 150 million companies because they have a level of scale, but they're not big enough that... Something weird happens to companies when they grow beyond 150 million. The problem is that if you're talking to a $40 million company, I can literally go and talk to the COO of the company, I can talk to the CFO of the company, and they want to do things, they want to change things, and if they really have a business problem, they're like, "Okay, I want to be this like $100 million dollar company. I need an ERP, I need this, I need that." They're not afraid to rattle the cage and basically rip out some WMS system or standalone systems that basically they have grown with.

But once you start growing beyond that 150 and you don't have a strong core, like for example 10 years ago, 100 million, 150 million company would not be able to afford a proper ERP, so what did they do? They had an accounting system separate, they had a warehouse management separate, and then they would... Okay, the operations guys like the warehouse management system, the accounting people like that accounting system, but there was a big disconnect, so they would spend a lot of unnecessary dollars trying to synchronize these two. If you go to a $500 million company and ask them, "Hey, we have a holistic solution that can really solve the problem for you." They're afraid of touching the operations guys. They're afraid in touching that. They'd rather spend more money, even though as inefficient, as inaccurate it is, just doing bandages between that accounting system and patching system. I try to stick to 5 to $150 million companies that want to grow to that half a billion-dollar companies. Then when we propose a solution, they're not afraid to basically go and do the right thing.

The second thing is with these kind of companies, it's fairly easy to understand what their books are like, "Hey, how much time are you spending on this?" Usually you can track that down to a particular general ledger account in your expense, "Hey, this is impacting this cost, this is impacting this salaries overhead." I can get that answer without a whole lot of BS, right?

Alison Cossette: Yeah.

Vish Puttagunta: I mean, there's no... It's not posturing. Once you go beyond that, it's like... You got to understand also their risk tolerance. When you're talking about a $500 million company, nobody wants to take that risk. It takes a very special leader to take that risk, right?

Alison Cossette: Yeah.

Vish Puttagunta: Only a CFO will be able to really say, "No, we got to go in this direction. If this works, it's going to be great for the company. If not, I'll get fired." How many CEOs can basically say that, "Okay, I'm okay getting fired if this vision doesn't work out," and actually still go ahead and take that risk? That's a challenge at the bigger companies I feel.

Alison Cossette: When you think about the ChatGPT initiatives and those generative AI applications, do you foresee that those large companies are going to have that same risk aversion or do you think that they're more likely to be, "It's where things are going, we don't want to be left out?" How are you seeing those kinds of applications across different enterprises?

Vish Puttagunta: Here's my bet. Enterprise guys have seen too many proof of concepts. They're tired of proof of concepts. They want something that actually just works. I mean, to be honest, I credit a lot to one of the CIOs that I worked with with one of the large firms. One time I built an AI model that was making about 300 to $500,000 in additional sales every week, and I asked a question like, "Why are you hesitant about sharing this data to your CFO?" The answer that he gave me was terrific. It changed my life. He basically said, "Vish, what's the guarantee that this will work again and again next year? Are you willing to take that risk? If you're willing to take that risk, I will basically go and connect you to the CFO or even the CEO of the company, but do you have those balls to basically put your foot down and take that risk?"

The reason why we're working at this level right now is that you got to make sure that when you... I would think the companies that will really succeed are not people that will say, "Oh, this model might work. That model might work." "Hey, we've proven this model across 15, 20 companies. It will work, and you know what? We will take that risk, but give us a percentage of the savings or percentage of additional sales." I think those are the kind of companies that will really succeed in the future.

Alison Cossette: Yeah. Got it.

Vish Puttagunta: Because, I mean, you go to a CFO or a COO, what are they going to know about Neo4j?

Alison Cossette: Yeah. Yeah.

Vish Puttagunta: What are they going to know about the difference between a graph-based algorithm versus a statistical algorithm? They don't know. It's just AI. Everything is ChatGPT for them, right?

Alison Cossette: Yeah.

Vish Puttagunta: But how do you break it down to a CFO? You basically tell them... They understand risk better than anybody else, so you go and tell them, "Hey, I can reduce this risk for you. You're spending this much money every year. I can reduce this to this much. I think I can reduce it to this much, and if I don't do it then I'll take the risk on that." Then the CFO knows how to compute the risk and fund that effort.

Alison Cossette: Got it. Anything else that you... Any other wisdom that you want to share with the audience before you go? Because you have so many good things.

Vish Puttagunta: I mean, one thing that I really struggle with is how do we... I mean, I think it's just a matter of time. In the end, we got to make our customers really successful. One of the challenges of a technical founder is the brand visibility and the sales is always a big challenge for a technical founder because one of the... I would say one of the things that a technical founders has to watch out for, if it's a sales founder with a sales and marketing background, that guy would've probably raised about $10 million by the time they even come here, right?

Alison Cossette: Yeah. Yeah.

Vish Puttagunta: Because they would not mess around. They would just go and spend 200, $300,000 a head and just hire those guys versus a technical founder will build a lot of stuff, right?

Alison Cossette: Yeah.

Vish Puttagunta: So we didn't raise a lot of capital, but our IP is worth 5 to $10 million easily. I think that's one of the things that we struggle with as a company. I would say from a technical founder to technical founder, if there's a technical founder out there that wants to do something, raise enough capital and have people on the board that are from the industry. That was the one mistake that I did looking back with my previous startups is that I did not have people from the industry on my board that have significant amount of shares in the company. When I started this company, I made sure that... The guys who started this company, I have people that actually are from the food manufacturing industry on my board.

Alison Cossette: Yeah. Yeah.

Andreas Kollegger: Good advice. I like it.

Alison Cossette: You can see why Vish was our guest today. Oh my goodness. All right, so at this point in our podcast we usually do something called tool of the month. Well, ABK and I will go first and if you have something that you want to share with the audience, we'll let you do that. Sound good?

Andreas Kollegger: We tend to go with technical things, like, Alison, I don't know if you have something on hand. I've got a... It's a bit of a Neo4j thing, though. Do you mind if I go with that first?

Alison Cossette: Please.

Andreas Kollegger: [inaudible 00:39:43]. So speaking of GenAI, this is a little bit of a bridge, I suppose. We've been doing lots of things with GenAI at Neo4j, of course, and one thing that's been worked on in the developer relations team for the last month or so is we've realized that for developers, of course, everybody wants to get into GenAI but you don't necessarily know where to start. If you want to use Neo4j with GenAI, what do you do? We now... It will be landing this month really, like by the time this podcast is out you should have them in your hands. We have these GenAI starter kits, so if you want to use Neo4j with any of the most popular orchestration frameworks within a GenAI, so if that's LangChain, LlamaIndex, Spring AI or Semantic Kernel even, you can now find a starter kit that uses Neo4j with that technology and just get a project going really easily.

One of the things I love about this is that while for sure and probably for the foreseeable future most of the work that's happening in GenAI is in Python and Python is dominating the overall story, this is still... Python, it's still LangChain, LlamaIndex, but then also branching out into Java with support for Spring AI. Semantic Kernel, if people aren't familiar with it, if you're on the .NET platform, you're not left out. If you're not a Python person, you can still do GenAI stuff using .NET using Semantic Kernel. It's a lovely framework. I recommend checking those things out. They'll be pretty great.

Vish Puttagunta: Sure.

Alison Cossette: Yeah. My tool of the month is actually one tiny little function inside of APOC, Awesome Procedures on Cypher, which is vRelationship. Being a data scientist, I always like to noodle and experiment and play with things. We know that within Neo4j relationships, especially if you have a really large dataset, they take up a lot of storage space, more than a node, and so what vRelationship allows you to do is it allows you to create a virtual relationship. If I want to just take a look at what that would look like for a visualization but I don't necessarily want to store it in the database because we're working on some dashboards for a community project right now, and so vRelationships are a way to show a relationship. Often it's sort of what I call a jump match.

For example, if I have a customer who placed an order and that order contained a product and I want to create a relationship among the products, so sold with, but I don't necessarily want to store it, I may not be using it in other times, you can actually use that and then take a look at, "Do we have, I don't know, closed networks or tightly connected networks of products and different things like that?" My tool that I was playing with most recently is vRelationship. I do have to jump in on one other thing, which is I just came from the first Microsoft Fabric Conference and it was amazing. OneLake on Fabric is just a really interesting way to bring together a lot of different data points, and for me as a data scientist anything that's going to help me with data engineering, getting back to what you were saying earlier, Vish, as a data science when you've got lots of different data sources, it's really interesting. We got some good Neo4j integrations there, but it was really exciting. It was a fun one. Any tools that you're playing with, Vish? It doesn't have to be super technical. It could be a website or a reference or even [inaudible 00:43:10].

Vish Puttagunta: I think for us, just adding that vectorization support was... I don't know when this got added, maybe it's been a year or so, but that is a big deal for us because now at this point of time, we can vectorize the datasets, paragraphs or entire PDF documents, and then we can store them in nodes. When somebody asks the questions, we can vectorize that question and see, "Hey, do a cosine similarity and get me the closest 10 nodes." Then we can give a very constrained answer, "Hey, only translate this information into a verbal response. That is pretty darn solid. I mean, I would say that was a very good move on Neo4j's part.

Alison Cossette: Well, that's definitely ABK's territory. We'll thank the vector team for you for sure. Wonderful. All right. At this point, Vish, we usually switch over to talking a lot about more specific Neo4j upcoming events and things like that. We know how valuable your time is, we don't want to keep you, but a very sincere thank you for coming out and joining us this month on GraphStuff and I'm really glad that you were able to make the time, so thank you so much.

Vish Puttagunta: Sure. Thank you so much. Thanks for having me. Thank you. Take care.

Alison Cossette: All right. Take care. Bye.

Andreas Kollegger: What a great interview. To your earlier preview of before we got the show started, what a great interview with Vish. It feels like you could go through any conversation with him and he'll have interesting things to say, interesting insights, and experience with it. Great guest to have.

Alison Cossette: Yeah. I was so happy when he was available because when I met him, he was just so compelling and he has so much deep knowledge about really impactful process assessment. What I really appreciated about what he said as a consultant, when you come in as a consultant... Here's the thing. I was always really curious how new graduates could come in and be consultants. When you don't have a lot of business experience, can you provide something that's more than general? There is... Even as technologists, and in many ways we're consultants when we come in with our technology, that intimate access to the process just brings such a deep, rich ability to analyze that I just thought was so fascinating, and what it compels me towards in many ways is from the business side, when you're running your business, like he said, as the CEO, the CFO and you're looking to technology companies or consultants or you're evaluating a technologist, the more you come to the table, the more open you are about your business, the more that technology is going to be able to help you, right?

Andreas Kollegger: Yeah. [inaudible 00:46:16].

Alison Cossette: We find that even just at conferences... We were just at Fabric, and the more time you spend with people and the more they share what are their pain points, what are their troubles, what's really happening, there's so much more that's possible in being able to help them solve problems.

Andreas Kollegger: Right. Right. I feel like sometimes... I agree with all of what you've been saying and that. Two points, I guess. The sort of fresh out of school aspect sometimes of your consultants, I think there's a fresh eyes, here's the latest thinking that is happening, is valuable-

Alison Cossette: True.

Andreas Kollegger: ... but it has to be balanced with the, "By the way, you know your business." And it's the balance of bring your business, deep intuition, and the experience of that with this fresh perspective. That's when consulting works super well, right?

Alison Cossette: Yeah.

Andreas Kollegger: I think that dynamic can be very powerful. Maybe the other aspect of that that you touched on, and I think Vish mentioned this a little bit as well, that for business... When you're in those kind of settings, you've got to basically come clean. You've got to be like, "Okay, let's have an honest discussion through this stuff." Don't try to paint a picture like, "Here's what we look at." We look at this and they're like, "Let's look at all the everything. Let's be honest about it, have an honest conversation. Otherwise, nobody's going to get anything good out of it." You don't go visit the doctor and be like, "Yeah, no, I feel fine." If you're going to go see the doctor, let them know all this stuff and then sort out what matters and what doesn't afterwards. Don't worry about it. If it seems like it might be relevant, say it anyway. You'll figure it out through the process of it all to your point about process, and that was Vish's great strength was through all this, was his very... What seemed like a very methodical way of thinking about any given problem. He was lovely.

Alison Cossette: Yeah. I will say in defense of consultants, let me just say to all the big consulting companies, if anybody's listening, we do recognize that the teams that go into your clients are not just new graduates. They're one member of a much larger team with a lot of years of experience, so not to... I didn't want to undermine or slag the consulting business at all. Anyway, that said, always a pleasure to be with you as well, ABK. Thank you so much. It's great to see you.

Andreas Kollegger: Yeah. Good to see you too.

Alison Cossette: It's been really fun watching all the work you're doing with Neo4j and GenAI. To that point, we [inaudible 00:48:42] going on in that space as well as others, so within the show notes of the podcast or on the video notes, just know that we've got the latest information and the latest articles that we've got going. We have actually right now a ton of live workshops. There's a lot of series of live workshops that are happening around the globe from London to San Francisco, so definitely check that out. If you're listening to this later, not when it's initially cast, I'm sure that that will still be true so you can always visit us at Neo4j and look at the Neo4j blog and our YouTube channel. We look forward to seeing you all next month at GraphStuff.FM.

Andreas Kollegger: Stay connected.

Alison Cossette: Bye all.