Welcome back to another episode of adventures in machine learning. I'm one of your hosts, Michael Burke, and I do data engineering and machine learning at Databricks. And I'm joined by my terrific cohost.
Ben Wilson. Yeah, just great code at Databricks. Yeah.
That's all he does. Nothing else. And today we are speaking with Sveks Sachtev. He started his career at Citibank in Deloitte and then moved to Bhutan airlines where he became a managing director, very prestigious position. And throughout his career, he's worked as a professor, as an entrepreneur, but his most recent project is Cantrak, which is Thailand's leading cannabis and hemp to seed sale software. Uh, and actually, sorry.
I misread that. Let me mark the clip. And throughout his career, he's worked as a professor and entrepreneur. But currently, he works at Cantrack, which is Thailand's leading cannabis and hemp seed to sale software. And just like with any industry, they leverage lots of machine learning techniques, which we'll get into today. So Shovek, diving right in, what are some of the biggest use cases you see for machine learning in the cannabis industry?
Hi guys, Shavek here from Thailand. By the way, it's 11.15 PM in Thailand. So I guess it's early morning where you guys are at. But back to your question. So machine learning, deep learning AI, I think at the moment, you can't go wrong with chat GBT or large language model in the application of chat bots. That's something which is in your face, by the way. So
Cannabis actually has its value, Shane. You have the guys who are growing. You have the people who are in the production extraction industry. You have the guys who are just the seller, the distributor. But I guess right now where I see most of it being used is at the dispensaries in Thailand. I think that's an opportunity, but I don't see it happening much. But I guess your chatbots or your
You said chatbots would probably be the best application of machine learning in the cannabis industry right now, speaking about just Thailand, you know, and how much we came, how far we came from the decriminalization.
Yeah. So I just wanted to push back a little bit. So LLMs, they can only talk to people. Is that really the primary use of ML of sort of interacting with customers? Or are there other things like harvest prediction, classifying whether a plant is healthy or not, or is LLMs really the only use case?
Yeah, well, so basically, LLM is probably something which is in trend, you know, so that's why I mentioned it first being like, you know, something which you read about it every day. You know, I wrote a blog about how to apply LLM models and to chat bots application as well, specifically targeted at dispensaries where they can provide information about the strains, you know, and maybe perhaps sell like those products which are legal for them to sell online.
because not all the products are available for sale. But yes, you can provide in-store experience information to the customer. You can even provide online information to the customer whenever they say hi to you on Insta or on Facebook. But you're right about how it will be used the same way it's practically used in other industries. So, forecasting, for example, but again, like
The amount of data which people are collecting is quite skeptical, which is why, again, I'm emphasizing on the LLM's use case more than on the predictive use case.
So if you were to look at applying something that's not specifically your end customer interfacing, but utilizing some of these new nascent tools that are hitting production quality.
production quality, you know, provided that you do things with the output. So a lot of the things that we're seeing on, uh, like hugging faces demo website where people are saying, yeah, it's great. We have MPT 30 B out there as an open source model that is pretty kind of close to chat GPT 3.5 and people aren't just wrapping that with an interface and saying, Hey, I'm going to create a.
you know, instructing, you know, GPT model or a chat bot that maintains state. Um, some of the interesting things that I've seen recently are like, Hey, I'm going to, I'm going to create a text classifier or summarizer, like tell this GPT model to summarize whatever somebody is saying and standardize it in some way. And I'm going to feed that into an image generation. So stable diffusion or something and do something clever like that. So you can.
then do fun things like, well, I'm going to take a whisper model along with a GPT model, like an MPT instruct. And then the output of that chain using something like Lang chain is then going to go to generate an image. So you can take your favorite music from Spotify, if you have a premium account, download it, put it in there, and then see what image it comes up with. When you see stuff like that.
and how people are leveraging, particularly with tools like Langchain. Do you think that's going to be revolutionary as a tool that growers would use? Where you say, hey, if you're running a farm that's growing your product, you're not hiring software engineers or data scientists to work on that farm, but you need technology in the hands of the grower that they can interface with.
Is that the sort of thing that people are thinking about in your industry or your company thinking about like, how do we give a tool where somebody can say, can take a picture of a plant and say, I think I have a problem here. What should I do? And then have it actually convert that in that language instruction into, you know, look up data as well as providing answer responses of here's the, the things you should do.
Yeah, I think you hit the point. You hit the spot right there, Ben, because there's something which we were working on internally, but we have not yet launched it to production nor testing environment. So I can't even call it beta. It's something which I was playing around with, to be honest with you. And I like how you talked about the use case of using large language models in terms of like chat GPT, for example.
And not just using it as a chatbot, but more or less using it as a text classifier, using it as a sentiment analysis or, you know, text or topic classification, because I tried playing around with that as well. You know, so what I did was because the growers right now, all the farms in Thailand, especially the cannabis and hemp farms, they're trying to go for those export grades. Right. So what's required from them is standards like, you know,
good agricultural and collection practices, GACP, GMP for manufacturing. And so it comes time for them to actually be very specific on what they do, right? So one of the pain points which the Thai growers is facing right now is what needs to be done, right? Like they have, they started growing indoor and that came in like all of a sudden, right? So people were not really prepared for indoor,
growing. It came in, everyone is doing it now. But as far as I'm concerned, the Thai growers who used to be underground, you know, growing it illegally, there's not many people who know how to grow it in a proper commercial indoor facility. So my point is, what we did was that we got these instruction manual, right, which comes from whether it's the LED provider, whether it's the actual facility designer, we shopped that.
into a storage. And we use Langshane to actually orchestrate the entire extracting the task required to be done from the PDF file and convert that into a set of instruction, which we have in Cantrak. So Cantrak has this feature called Work Schedule. Work Schedule is pretty much your to-do list. So
The master grower is supposed to be keying in a template of what needs to be done. And then he'll assign it to the growers on field, right? And then each day they'll take that task and we have this really neat app called the growers app where it runs on their phone and it only has the bare minimum amount of things which the grower needs to know, right? And then what we're trying to do is that, so to automate this task, we're trying to use a chat, GBT base or a, you know,
all those models available in the hugging phase right now, right? To be the means of transforming, extracting and transforming tasks from an instruction manual into a to-do list and assign that to a grower. And it did work pretty well, you know, because in our application, we are trying to tell the grower exactly steps by steps of what to do, right? Even with like what equipments they need to go and get from the store.
and what supplies they need to go fetch from the storage room. So those things are actually classified as well. So my prompt would be something like, with this instruction manual, extract things to do and classify all the equipments and supplies required. So each item, each work task will have a description of what needs to be done along with a tag of supplies and equipments. And that can then be integrated with our system. So that is something which.
You know, I've been building on and off as well in the background.
Yeah, it's a really intelligent way. Everybody that I've talked to, um, now that I'm on the other side of the fence of not building app, you know, applied uses of this stuff, but more thinking of how do we build the infrastructure to make that easier? And we talked to, we still talk to a lot of customers who are working on things. And I'd say a lot of people are in that sort of just playing around phase. They have. It's like they.
They now see this technology and they're trying to think up ways to apply it to their business, but all the one, the minority are the people that are talking about exactly what you're talking about. It's like, here's a problem that we need to solve and this makes our implementation and you know, what we need to do orders of magnitude easier and cheaper and faster. And those are the people that are.
I wouldn't say anybody's really in production yet for this stuff because it's still so new and there's a lot of wrinkles to iron out.
Like looking at the technology as just a tool to solve actual problems that you have instead of inventing problems to solve with a new technology is really smart. I do have one question though that is highly specific to the country that you operate in. So for listeners out there that are not aware, the Roman alphabet is not used in Thailand. It is...
It is a thing. You go over there and you will see English on signs everywhere. A lot of people know how to read certain things. And I'd say English is the second most frequently spoken language in Thailand. But when you go into rural areas, you're definitely not hearing English ever. So a farmer is going to speak Thai, and they're going to read Thai. So.
Yeah, definitely. No way. Yep.
When you're building applications that leverage things that are built in the West in English or on, you know, Roman based alphabets and also, uh, Roman based language structures, because all of these language models that are trained, they're trained on a lot of English or romance based languages. So it might be French, you know, it might be, you know, Italian or something.
But the fundamental structure and conjugation and the way that sentences and meaning is conveyed in a phonetic language and the fact that you don't have this sort of standard Roman based language structure, how do you fine tune something and create an interface that a grower would be able to use?
Is that a big challenge and how do you tackle something like that?
It is. It is. Well, you know, for basic things, it will just be creating a different JSON file for a language pack translator. If I can do that, then that's the first thing we do. I mean, our dev teams also, you know, they don't speak Thai. So we have teams, we have remote teams in India as well. They don't speak Thai. So we're facing a lot of good challenges. It's quite fun. You know, like our meetings are quite fun. You have like Thai, English, Hindi.
all jumbled up in one conversation. That's pretty fun too, but that's the culture, I guess, of working remotely. As for the Thai market, the actual user, right? You'll be surprised to find out that many growers in Thailand are not Thai. So, you know, some way or another, like that's being handled by themself. But yes, today I literally had the challenge of trying to explain to a Thai grower what a grower's app is and how they can move, how can they actually...
update the status of a plant being dead or being classified as a waste, because it's literally there on the screen, but it's just like a lot of challenges to get the Thai people, the Thai growers especially, to know how to make it work. And I'm not saying anything bad here, but the technology penetration rate in Thailand is very, very high. Like people get their hands on technology, but the actual adoption rate, like the adoption in this case,
putting technology to use, right, is not so high. Right, so that comes itself with a challenge which we're facing every day. So to build something which is Thai facing to the Thai audience or the Thai user is very challenging, especially, you know, when all the latest trend and technology are actually based in the Western world, if I can say that.
It makes sense from time that I've spent over in that country many decades ago. If you get out from the major city and get into farmland area, you start seeing that people are doing things in ways that you would never see in Europe or North America about processes of farming. And you look at what they're producing, like food, right?
But you look at the yields that they have and how they're able to get something that spectacularly good because of land management techniques, water management techniques, they're still using tech. Yeah, but they they're using these techniques of, you know, animal husbandry on the farm and just ways of adapting to the cycles of nature.
It's in the soil.
like, hey, when is the monsoon season coming? What days do we have to do these tasks? How do we burn our fields down so that we replenish the soil? This is stuff that's been in their knowledge base for a thousand years and they haven't really deviated from it because it just works. So I could see how adopting technology in a way that it's almost like you have to ride this fine line when you're providing advice to somebody who...
already does something exceptionally well, it has to be pinpoint to solve an actual problem that they have instead of telling them. And this applies to any applied ML. It's a sort of like a meta problem. If you're building something that, that counteracts what somebody's understanding of reality is, and you're saying, no, you're not doing that right. This is a different way, a better way of doing it. If that suggestion is actually wrong and they know it's wrong, they just lost faith in everything that you're trying to tell them.
Exactly, exactly. Yeah.
So that's a tough challenge to be a part of.
Yeah, we try something like that as well, using deep learning to do image classification of diseases and stuff like that. But then when we get it wrong, they just lose trust. I told them as well that this is still under process, it's under trial. But yeah, in the end, we managed to get up in terms of partnership with another company. So that's pretty much solved now, for us at least. But believe it or not, you still get like,
growers who literally take images of things which can't be classified using deep learning. So you know, they were supposed to take images of like, let's say the plants, right? But they, instead they took like the images of the clones or the seedlings, which in the end, like, you know, not much classification can be done. So we learned from this mistake and then so we put in this rule-based condition where we've either checked the images first before sending it for a...
API call, or we just do a high and fast block for them to not send these images for inference at our servers in between the first few stages of the growth stage of the plant, because the images coming up will not be of any use for inference.
Interesting. So do you do some sort of edge deployed model if you want to do it on an app saying like a really lightweight YOLO model saying, is this like, what is this? And if it comes back saying it's a plant, then let it go through. And if it's like, this is a person's shoe, then. You know, alert the user like, Hey, this is not, not good.
We do use something similar to that sort of for stock taking. So like the pre-harvest activities, right? The master grow would then instruct the grower teams to go and do a stock count of all the plants. This is prior to harvest and defoliation, right? So they will go around and try to scan each plant for counts. But then we also ask them to put in the lot number, the strain and stuff like that. So things can be cross-checked in the backend. So we know that there are no discrepancies
What is in the system? What is in Cantrak today? And what's actually in the physical world, right? That's when we do a stock count and then we do it one more time when it gets into the drawing room Just to keep tally of all things
Cool. I have a question. Um, so sort of going back to a topic that we were discussing earlier, where this is a very nascent industry. There's a lot of, uh, opportunity for building new things. Um, how are you thinking about prioritizing what tools add value? Because theoretically, and I was just looking at Cantrax model, they, you guys are very vertical based. You don't, you don't specialize in one sort of area of the growing process. You go all the way from basically seed to distribution.
And so because of that, you have many possible applications for ML. So how do you think about within the entire organization? How do you think about targeting specific models and specific use cases? Cause I would think it would be overwhelming.
So the mindset which we are trying to lay as a foundation, I mean, with the team right now is we're trying to move away from just Cantrack being for the cannabis industry to for Cantrack being for the economic crop industry. To answer this question, which people normally ask is that, can we use Cantrack for crop A, can we use Cantrack for crop B? And the answer is, if there is value in terms of what you're doing, right? Then yeah, by all means use it. So with that mindset,
we tend to then move away to other crops where we can play around with more data. And with more data, that's where the actual idea pops up of how we can actually use ML. But then again, because we're a startup, right? And everyone is just busy building. And it's very difficult for me to actually make my way through the team and say, hey, guys, let's do a quick challenge here. Can we use...
this particular package or this particular ML model to solve such and such problem, it's going to be quite difficult at this point. Because what you say is correct. Cantrack is a C2Cell platform or C2Cell system. So we need to have every single feature in that entire ecosystem. Otherwise, they will not choose Cantrack. It needs to be a one size fit all. It needs to be a one stop solution for the customer.
That's our target, our priority right now. ML, yes, we do. We'll try. We have put it on a trial a couple of times. And sometimes it's very useful for the developers, as well as the end customers.
Yeah, that makes sense. It seems like you should. Well, just with any ML project, you need a lot of infrastructure around data before you can even apply a model. So it makes sense that you guys would be targeting it. It also makes sense that you guys would be expanding into other industries. Um, for instance, I have a few friends that work at John Deere and the use cases of ML in that organization versus Cantrack there, they're night and day. Like John Deere has, uh,
Basically one, they have entire teams and entire departments focusing on automation of tractors and, uh, sort of Cantrak is looking to sort of. Go a different direction and empower the farmer to be more efficient at sort of the user level. Uh, so it's a really interesting device.
I mean, I got to say from a product perspective, this is pretty impressive to me. So there's already companies that specialize in this for food production, very large companies that do supply chain management for growers and getting stuff to the market. We're talking about tomatoes, right? Or corn. If a startup were to come in and say to the farmers that are doing that,
and the supply chains that are doing general food distribution. If somebody went to central Valley, California and said, Hey, we got a solution for you and it's out of Palo Alto, California, and we're going to revolutionize this. Nobody's going to use it. Or if you attempt to use it and attempt to sell it, those established companies that already do all that stuff that have been around for 150 years, they're going to drink your milkshake.
They'll reverse engineer whatever you're doing, or they'll just put you out of business, sue you, whatever they can to just destroy you. But going after the one thing that is very profitable in the world today, in the 21st century, that those established companies don't want to touch because of the negative association that exists in the public consciousness for like, Oh, this company that
that handles our grocery store food. Why are they doing marijuana to dispensaries? You know, extremely conservative people would, you know, have a problem with that. So it's like you attack, you went after the one part that those established companies aren't going to touch, but then you get serious and successful in this one area and now you're, you are poised to actually go after those other areas. That's very smart. I just wanted to say that. It's like.
What a brilliant business decision. It's pretty cool.
it will actually corner to think this way because, and how it is in Thailand is pretty normal. Like you have the government, the political risks. Thailand's in terms of political risks is probably ranked as one of the top tens, I guess. So to actually put all my eggs in one basket and convince the entire team, because we have other products we're working on apart from Cantrack, we also provide consulting businesses to keep the company afloat of course, but you know,
There was one point in time where I had to convince the team and other stakeholders that, please trust me, I still got this. We will add more customers to Cantrak as a SaaS. So there were questions being thrown around. People were very skeptical about the cannabis market, about how they are able to finance these expenses or investments in software and track and trace, especially when compliance has been thrown out the window as of last year.
We thought before decriminalization, when they open up, we still had a foothold in the market where we can actually claim that we have the compliance platform, you better use this. Because based on my background in consulting in Deloitte, I was always taught that compliance is the easiest thing to sell. If you have a solution for that, then it's a market-ready, market-validated solution. So we were actually cornered to think this way,
use Cantrak for other crops. And I was also encouraged by many people that if we can nail the cannabis industry in terms of track and trace, then we should be able to do it with other cash crops as well or other commodities as well. Because cannabis being narcotics or classified as narcotics in many countries, or just recently removed from the narcotic list in many countries, it's classified at medical grade or medical cannabis. You do require to do a lot of.
tracking and tracing. So that's the mindset that we have a comprehensive system, yet it's not that ERP-ish difficulty level of usage. So people don't actually have to go lock themselves up in the toilet and cry when a system like SAP, these things are being implemented. So it's a more user-friendly, a more scaled down for cannabis companies. And what we usually tell them is that it's not just about compliances, it's not just about
track and trace, just do it as a discipline and you will get other things along the way, like quality insured or cost tracking. Cost tracking is one very important thing which people in this industry here in Thailand is just overlooking or just not paying much attention to it at all. And if you use CanTrack or any proper track and trace system and you do proper data entry for all your supplies, right?
And then if the supplies are then associated with activities carried out by the grower, and once that activity is completed, or once that supplies has been issued out from the inventory, you know, you're tracking your cost of the activities along the way as well. So that's the byproduct you get from track and trace. That's how we feel.
Are you guys using anything like blockchain in your stack?
We did one project. We didn't include blockchain as one of the key things in our tech stack, but we had one project where we were doing it for a research university, and they were trying to prove traceability on blockchain. I was lucky enough to have GBT as my assistant in coding, so that was pretty easy. That was this year. That was April this year.
We were putting all the key critical data on the chain, right? And we were using a blockchain as a service. I'm not going to mention the names of the blockchain we used, but it was a blockchain as a service. So it's just API calls. Very easy.
So if you were to start looking into similar industries, or you're saying, hey, we have the compliance stuff all set up on our platform and we know how to do that, we've gone through the growing pains. Is there anything stopping you from doing legally regulated narcotics? You're like, hey, opium growers, we got you.
That's not really a mom and pop sort of thing. Uh, that's usually industrial scale, you know, growing, but is stuff like that on the horizon or is it more, Hey, we want to diversify to, you know, food markets in, in Southeast Asia.
Yeah, I think the first one which you mentioned was more something which is closer to what we're looking at, right? Looking at other crops or other plants to manage. We did have a few conversations talking to people who were growing tomatoes, salads, even rubber, right? Because yeah, rubber. There is a need for, there's no need, I mean, there's not a clear need for track and trace, but there is a clear...
need for traceability across the supply chain. So it's the same thing with cannabis. Once you take out your harvest, then you ship it out for extraction at third-party facilities. If you don't have your own extraction facility, then you need to be able to still keep track of that particular batch number or lot number when it returns from the extraction facility. Because then when you sell and if anything goes wrong, then you can still trace back, even though it's being sourced externally.
So I think that is a common question across all the trades industry. So rubber being exported to other countries needs to be tracked, how it's actually being purchased. And then you can get into the nitty gritty of, is it fair trade? Was it actually procured from the right sources of grower? Are these guys really Thai? Are you sure? So a lot of questions are being asked.
Yeah, can track in that angle will be able to suffice and then that's the whole pitch of you know can track not being cannabis tracking per se but more like can track. Yes we can track. That's the whole point of it I guess. And all these things were not thought off from day one. It just came up to us.
It almost seems like another industry that is agriculture in nature that people really care about the sourcing of, at least from my limited understanding, is coffee. Is that something that would fit into that platform? It's like, hey, I want to know, you know, there's coffee aficionados out there that are like, well, I want single source. I want to know what farm this came off of. I want to know all the details about the roasting process. So you can track all of that in that software.
Um, so if you have all that data, let's say in the future, two years from now, you're now being used and, and your, your SaaS product is, is available for 40% of all cannabis growing for recreational use globally, and let's say 10% coffee use for growers, that 10% coffee use is probably a thousand times more data than, than the cannabis data, just due to the volume. Um,
How do you look at rich datasets like that where you have insight into the entire process of from when seeds arrive at a facility, they start germinating all the way to you can track that individual plant seed to a point of sale location anywhere in the globe. With that amount of data, would you see?
Do you see the future of profitability for the company being the data set itself rather than the SaaS model?
Yeah, definitely. You know, like building a big data out of what we have would be tremendous for this industry, right? The insights, the amount of knowledge it has, it's going to be enormous. And we do have experience handling this sort of data. We've had consulting projects for many poultry businesses where we saw their business growing, literally growing. It's a listed company in Thailand. So what they did was actually, it was just a factory.
where it was just a slaughterhouse to begin with. And then they expanded into the retail business. So we saw them from the days where they have only a few shops and then it started to grow and month on month grow and so on, 10X, 20X, literally, where now they have 100 plus stores across Thailand in the space of three years. And that also started during the middle of COVID. To actually see that data grow to us.
For that sort of a project, we actually had to build a data platform on cloud so that it can then be digested into S-Tree and then perhaps connected to Tableau or Power BI for further analysis of KPIs. And the insights which we get from those sort of a project is enormous, especially with like
Sales trend and just a simple things like forecasting your inventory is it's already giving business the value in the returns They're required to invest in such project with cannabis I guess we can do the same but I do agree with you that it's a it's a Pareto law right at 80-20 balance where you might have 80% of Cantrax user being cannabis companies and that 20% being that coffee Manufacturer and that 20% itself would be enough to fetch us 80% of that rich data for us to do
amazing things with it later on.
Yeah, that makes sense. Just to sort of communicate the scale and the potential of this industry, we were chatting a little bit before we started recording about how you got a New York Times interview and we won't spoil the New York Times topics, but yeah. Yeah. So one of the topics that I'm sure they were interested in is sort of the potential of this industry and the potential valuation over the next five, 10 years.
Yeah, it's not yet published though.
What are your thoughts on how big this industry can scale specifically in Thailand, but also maybe in Southeast Asia or even globally?
Well, I think Thailand being the early mover, right, is probably going to be looked at as a role model. We've seen already people from Malaysia, people from Indonesia, people from across the world looking into not just investing to Thailand, but learning from what the Thais are doing here, you know, especially Malaysia at the angle where Thailand is treating this as medical cannabis and CBDs from hemp as well, is being studied carefully by.
Malaysian governments. They probably plan to open up very soon in the future. Thailand would probably be looked up as a model of roadmap in terms of what steps they take to legalize it and make sure that it's under compliance.
If you ask me, I guess it's going to be growing. It's still a... There's a lot of uncertainty surrounding this industry because of how it's being linked heavily to politics. That is something where... That's the only roadblock which I see keeps it from progressing ahead. It should be further down the line than what we see today, had it not been because of the turmoil which we're always facing.
Do you have any, yeah.
Do you have any idea on a number in terms of dollars? No.
It's difficult. Yeah, I was asked that question too by the guys from New York Times, but it's a very difficult number to find as well because, you know, although when you open up business and you have like free trades coming in, you have foreign investments coming in, it will just be classified as a business in agriculture, right? And then because it's not illegal anymore.
you can just file for a license. So you just have to tell them that you're growing cannabis and you're free to grow whatever you want. It's very difficult. Yeah, I can't put a number to it. It's gonna be very misleading. But every now and then a really huge farm just pops up. Trust me, that's how it is right now. It keeps on getting bigger and bigger and people are complaining that there's a huge amount of oversupply in the market, illegal imports and stuff like that as well.
Got it, yeah.
I mean, geographically, it's uniquely seated in a place where that plant grows really well. Pretty much any plant besides potatoes grow really well. And then if you want to grow potatoes, grow up North in the mountains. But you also have a sort of a culture of people from my experience who are very innovative and are willing to do.
whatever it takes to get something done. It's a very unique spirit of everybody that I've ever met from that country. Um, with regards to, Hey, even if we don't have the latest, amazing technology, we'll make it work. And they do. And, you know, when you have a, an ethos, like a work ethos like that in an entire culture, I think it's very interesting when an opportunity arises for
proliferation of something and being a first mover and being that model to other countries in the region or just globally as well. Do you see that as something that's going to shift the tide? If you say historically the government of Thailand was not very understanding of people who were doing things like trafficking marijuana.
Penalties were very harsh, jail terms were very long. Some people got the death penalty. And when you see that reversal of saying, actually, this makes economic sense for us to do and we're gonna decriminalize this and make it so that people can actually do this, do you think that's gonna change the political viewpoint of a lot of other countries that may not have been quite as conservative with regards to drugs, but also the economic turn that.
that people are going to see 10 years from now in that country. It's like, Hey, it's not just a tourist place. And look at this massive industry that we got going. It, it sparks of, you know, the transformation post World War II of South Korea, not post World War II, but post Korean war, where there's this revolution that happens in technology and all of a sudden they're the world leader in producing consumer goods.
It's still a taboo, I would say, today. People are still looking at it as a drug. But I've been hearing from a lot of parents that now they treat their kids better because the kids are actually now growing cash crops. As to before, the kids are actually growing something which is going to be consumers drugs. So that kind of changed as well. But yeah, I guess if Thailand can do it.
And Thailand being a Buddhist country, being a very strict country, and heavily relying on tourism industry, and we saw how badly it crashed due to COVID. I guess this may be the next big thing. If we do it properly, if the foreign investments in Thailand are being taxed properly, if the Thai grows are being promoted properly, then I guess it would be an up and coming industry definitely.
when you usually see this economic value from Thailand, it will prosper. And other countries surrounding it, which we're considering, it's like when your best friend has a new toy, a nice shiny car, or say you want it. And then now it's not expensive anymore. So it's not looked at in a negative point of view anymore. It's now a good thing to do. So I guess.
With the ever-changing world, it can be accelerated if Thailand happens to make it big and successful in this.
It's got to be exciting to know that you're working within an organization that is at the forefront of something like that. It's trying to dodge the political side of things, but saying technologically we're going to apply things that are very advanced, exceptionally advanced to promote this.
Yeah, that makes sense.
It still has that new smell to it, that new product smell. I was running another product. I'm still running multiple products at the same time, believe it or not, but not such a hands-on role as what I'm doing with Cantrak. But for example, we are still running this platform called ServiceMine, which is a building management, preventive maintenance, corrective maintenance. When things go wrong, you can just
create a ticket through your phone. Image classifications are being used to detect the category of the problem so that when you classify it, you don't have to go through that list of thousands problem category. It will just be slimmed down to just a few because of the images it has classified. That has a sense of that smell of competition. You're faced with heavy competitors everywhere, right? And you're supposed to try to make it true. But with Cantrak, there's still that...
Yeah, we do have challenges. We do have competitor. So at least we can rest assure that we're in a market where people are willing to compete. So there must be something in there, like startup textbook stuff. But then, yeah, it has that, you still have that new smell to it. When you open up a box or a product is new, you have that smell. I still get that with CanTrack. I still feel it's a very new, exciting industry. And what we're doing right now today is.
It may not be like a new innovation every day. Like sometimes we feel we're just building a data entry tool. You know, it's just forms after forms of data entry, of analysis. But then when it makes a difference, I guess that's when we actually see the value in it. And that when the companies here, the cultivators, the cultivation facilities here can obtain certificates and standards required for them to export, I think that would see us being very, very successful.
and happy at what we have achieved together with the customers.
It's sort of ironic that you guys are using LLMs as sort of this nascent technology as well in a nascent industry. But I was wondering if you guys are thinking about like what's on the forefront for Cantrack and ML other than the use cases we talked about? Are there any things that you're excited about on the horizon?
And you'll be.
Well, I think to be able to help the growers, especially because that would be a revolutionizing industry. The Thai growers, the language barriers, and also how the growers would then be running or positioning themselves in a company which is invested by a foreigner, running by a foreign master grower. And if we can use them out to actually make a difference in their life, then I think we're doing something correct.
So even if it's just as basic as using land chain to actually extract something from a PDF file, feeding it to a vector database, and then fetching it at real time, being like a advisor on how to use a system using chatbots, or even classifying topics, if it helps the growers or if it helps the Thai industry, then I think we're doing something right. You know what I mean.
With a rich data set that you could potentially get with, you know, if you have half the people in the country that are using your app, that are in the process of growing, and you're doing image classifications for pretty much every potential problem as well as every potential success for this plant species. You have this.
Yeah, we're doing that, right? So we, yeah.
data set that you could be storing and using for all these other potential use cases with that tracking data. And if part of that supply chain prior to going to a dispensary goes for testing and they say, hey, in order for this to safely be sold in country X, we actually have to get full lab testing of this to determine cannabinoid content, ratios of these different compounds.
If you're making CBD things out of it, it has to be analyzed in a slightly different way. But you have that link between which seeds were used and how it was grown and what were the actual growth conditions. What was the moisture content of the soil? What was the pH of the soil? What fertilizer was put down? You basically are running.
Yeah, how it grows. Yeah.
It would potentially have the results of an uncontrolled, but fully documented massive design of experiments, uh, as part of your data set. And you could basically sell that data to say, here's the optimal way to grow this thing.
and then have.
Yeah, now I'm thinking of a product, but you could.
Hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm, hmm
you know, leveraging an LLM with something like, Hey, I want you to distill all of this information that we have about these 600 different strains. And here's the successes that we had find the optimal conditions of what we should do for, this is where I'm trying to grow this. Here's my altitude, you know, rainfall data, sunlight and everything for the year. What should I do? And it sits there and goes and retrieves that data and
Yeah, yeah, autopilot.
Yeah, it gives you like the instruction set of like, here's the probable best yield you're going to get for the best amount of profit. It's pretty interesting.
It's ideal, I guess, in terms of the persona we're face-off with today. Like, you know, customers with money, investors with money, but no knowledge of how to grow, can just launch this app and choose a strain. And then it will just fetch like the optimal growth stages plan. Because whatever you have to said, we have it in the system. So we have the planning function where you can actually plan and, you know, create like planning variables of when it will be.
We will give them a number of expected harvest date, but then we will have the plan when they start growing. They will be forced to put in those data because it's a track and trace system. So we have the plan, we have the actual, we have the tasks applied to each of the roles and what needs to be carried out for what strain. And once, and we will have the yield because they're supposed to put in the harvest data. We have the production system so they can actually.
go through the trim, the dry, the dry, the trim, and then the curing process before packaging it, and the price because we have the sales system. So yeah, what you say pretty much is like another product can be built on top of this. When, and that's a big if, we get like a lot of users, it would be that in your face use case where if we are actually skipping, it would be a really dumb thing to do.
potentially very lucrative.
And if you get enough, you know, market penetration into different regions, you know, all you would need is just a couple of growers in the United States. Uh, and then a couple in Canada, you know, wherever they're like, people are growing this stuff. Uh, I'm no expert, but, um, it'd be pretty interesting.
And because most of the growers in Thailand are now growing it indoor, right? So that That external factor is ruled out. So we do have a recipe if that day comes then I guess we'll be back in this podcast Discussing about what's being done But but yeah That would be nice
So I guess it was back in the days before, before legalization that people were going out 20 miles into the middle of the jungle, trying to avoid tigers, uh, in order to plot the, you know, plant their little plot there that they would then illegally sell. But now you can do it in a small town in a, in a warehouse.
Yeah. But I still have this very interesting customer I'm working with right now where they're growing it organically. So no, no rock wool, you know, no non soil products, just purely soil, like living stuff being used to grow cannabis. And the beauty of it all is that they're also using or, you know, paying the tribal people in the mountains to help them with this.
It's something which I told them that they should do a documentary off and they said it's something which is already in the pipeline to build a documentary out of this.
that super cool. I would watch that.
interesting. That's the Hmong people, right? In the near...
Yeah. It's the Korean, Korean people. Yeah, Korean, yeah.
tribal people with cannabis in their roots, I think.
Cool. All right. So I know we're coming up on time and I know she's like, like 4am for you. So we should, we should probably let you go to bed at some point. Yeah. No, we appreciate you staying up.
Right, it's 12, yeah. Yeah, it's past midnight. Yeah, that's why I was feeling I was a bit... All right. Okay, awesome, awesome. Yeah, yeah.
Yeah, you feeling alright over there?
Okay, cool. Well, I will quickly wrap and then we'll kick it over to you. If you have any, uh, plugs or things to say. So today we talked about machine learning in the cannabis industry, specifically in Thailand, it's a very nascent industry and there's lots of opportunity. And that has both perks and it also has problems. Um, but some of the applications that can track specifically is using machine learning for is LLMs for chatbots, LLM for to-do list creation, LLM for text extraction from PDFs.
And they also are leveraging a tool called grow doc, which does sort of image classification for the health of a cannabis plant, and you can let your mind go, go wild with all the other use cases that they have, because they have this vertical system where they go from seed to e-commerce, so there's lots of use cases for ML then a couple of tips. Uh, we've talked about this many, many times, but trust is really hard to build an easy to lose when you're building machine learning models, and especially with non-technical people where you're.
trying to disrupt an industry and also bring technology to that industry. Uh, it can be really hard. So getting it right the first time and staying right is really important. And then for a Deloitte ism compliance is the easiest thing to sell. And then the final thing that I wanted to sort of highlight is the power that data can provide. Um, basically if you collect a ton of data and you don't quite know what it'll be used for, it's still a really good idea to collect it. Storage is cheap. And.
Having that data to look back upon is really valuable because you can build products off of it. And at the time of collection, you might not know what it's used for, but again, because collection and storage is cheap, it's really, really powerful. And so I suggest many companies try to do this as well. So, Chyvek, if people wanna learn more about you or can track, where should they go?
So you can either add me on LinkedIn. It's Shafak. My name will probably be here in the podcast title as well. Or else you can visit cantrack.tech. That's can track, no CK. So it's cantrack, T-R-A-K, dot tech. And we'll be launching a new version of the website very soon.
Sweet. All right, well, this has been a lot of fun. And until next time, it's been Michael Burke and my co-host. And have a good day, everyone.
See you next time.