1 00:00:00,000 --> 00:00:10,840 The strategy that I would love to do is you would basically mix some of the tried and 2 00:00:10,840 --> 00:00:17,520 true genetics like the MAC with some of these crazy land races that people are discovering 3 00:00:17,520 --> 00:00:24,280 in Southeast Asia or people are growing cannabis. 4 00:00:24,280 --> 00:00:29,480 One, you want to make sure you get your seeds from a legitimate vendor. 5 00:00:29,480 --> 00:00:36,720 People are growing cannabis all over the world from Afghanistan, Pakistan to Southeast Asia 6 00:00:36,720 --> 00:00:41,920 down to South America, plus all over the United States. 7 00:00:41,920 --> 00:00:51,360 Some of the best seed vendors are right in California, right in Humboldt County. 8 00:00:51,360 --> 00:00:58,640 And then there's the whole thing where you could do feminized or unfeminized seeds. 9 00:00:58,640 --> 00:01:08,880 And the more I looked into that, I don't want to harsh on anyone's seed business, but basically 10 00:01:08,880 --> 00:01:14,040 definitely the big time producers, they just go straight through the feminized seeds. 11 00:01:14,040 --> 00:01:15,280 They just don't want to mess around. 12 00:01:15,280 --> 00:01:21,920 They just want to grow a bunch of flower and just female flower and be happy with that. 13 00:01:21,920 --> 00:01:25,280 I'm more interested in genetics and whatnot. 14 00:01:25,280 --> 00:01:30,320 And so if you're breeding, you may not want to get feminized seeds, right? 15 00:01:30,320 --> 00:01:35,640 Because you need some males in there and all that jazz, but we're going down a whole can 16 00:01:35,640 --> 00:01:36,640 of worms. 17 00:01:36,640 --> 00:01:41,340 But anywho, enough seed talk. 18 00:01:41,340 --> 00:01:46,240 So for those of you who are new, Grant, my name is Keegan. 19 00:01:46,240 --> 00:01:47,760 Started a company, Canlytics. 20 00:01:47,760 --> 00:01:52,760 So got into the space first by helping out labs that test cannabis. 21 00:01:52,760 --> 00:01:58,960 And now we're just trying to organize and make cannabis data accessible. 22 00:01:58,960 --> 00:02:01,340 So it's sort of all hands on deck moment. 23 00:02:01,340 --> 00:02:04,120 So happy to have you here. 24 00:02:04,120 --> 00:02:06,160 Thanks. 25 00:02:06,160 --> 00:02:10,200 I'm happy to be here. 26 00:02:10,200 --> 00:02:12,520 I'm just kind of curious about the space. 27 00:02:12,520 --> 00:02:17,840 I'm a pretty young data scientist and was honestly just wanting to learn more. 28 00:02:17,840 --> 00:02:19,200 Too cool. 29 00:02:19,200 --> 00:02:23,200 Well, we've got a real interesting project at hand. 30 00:02:23,200 --> 00:02:29,000 At the moment, we're basically wrangling data from certificates of analysis. 31 00:02:29,000 --> 00:02:34,520 So basically, cannabis has to get mandated quality assurance testing. 32 00:02:34,520 --> 00:02:41,720 And while that can be a sore point for a lot of producers, what I try to point out is this 33 00:02:41,720 --> 00:02:45,080 is also putting a lot of data in your hands, right? 34 00:02:45,080 --> 00:02:50,360 These are rigorous tests, right, and you're getting all the compounds in your products 35 00:02:50,360 --> 00:02:51,360 measured. 36 00:02:51,360 --> 00:02:58,040 So instead of just leaving that data on your table, it will take full advantage of it. 37 00:02:58,040 --> 00:03:03,320 And so basically, people have all this data in PDFs. 38 00:03:03,320 --> 00:03:05,440 Or sometimes it's on the web. 39 00:03:05,440 --> 00:03:15,280 But it's amazing how much data can get locked up in a PDF and get rendered useless. 40 00:03:15,280 --> 00:03:21,840 And so I've seen it where people will have hundreds, if not thousands of these COAs, 41 00:03:21,840 --> 00:03:27,960 and they're just saved on their Google Drive somewhere and they're not doing anything with 42 00:03:27,960 --> 00:03:28,960 them. 43 00:03:28,960 --> 00:03:34,040 So basically, we're building tools here to unlock that data. 44 00:03:34,040 --> 00:03:35,100 And it's open source. 45 00:03:35,100 --> 00:03:42,160 So I'll show you today how, one, you can go pick up these tools and make money with them, 46 00:03:42,160 --> 00:03:43,160 right? 47 00:03:43,160 --> 00:03:55,440 You can go talk with labs and laboratory software providers and help them out and help out retailers, 48 00:03:55,440 --> 00:03:56,440 help out producers. 49 00:03:56,440 --> 00:03:59,880 So the sky's the limit here. 50 00:03:59,880 --> 00:04:10,600 And anyone have any thoughts, comments, questions before we just dive straight into the fun? 51 00:04:10,600 --> 00:04:15,280 One curiosity that I have is just generally how much analytics is actually happening right 52 00:04:15,280 --> 00:04:18,040 now in the cannabis space. 53 00:04:18,040 --> 00:04:22,720 Is data analysis or data science a pretty hot topic right now, or is it still kind of 54 00:04:22,720 --> 00:04:23,720 brooding? 55 00:04:23,720 --> 00:04:29,080 John may have a thought or two on this. 56 00:04:29,080 --> 00:04:37,280 My thought is there's a supply, but I think the demand far outstrips the supply. 57 00:04:37,280 --> 00:04:41,520 So it also can be a tough sell, right? 58 00:04:41,520 --> 00:04:48,040 Going to these producer processors or retailers, I think a lot would benefit from analytics. 59 00:04:48,040 --> 00:04:52,360 But their first question is going to be, what's my return on investment? 60 00:04:52,360 --> 00:04:58,460 So if you're not able to clearly articulate what their return is going to be, they probably 61 00:04:58,460 --> 00:05:04,760 won't even entertain a conversation with you. 62 00:05:04,760 --> 00:05:10,400 So like I said, some of the more savvy players are realizing this is something that they 63 00:05:10,400 --> 00:05:14,680 need, and so they're trying to find analytics providers. 64 00:05:14,680 --> 00:05:19,440 And now, as I said, I think there's just a very, very small supply. 65 00:05:19,440 --> 00:05:25,520 So if you're interested in bringing your skills into the space, I think there's a demand for 66 00:05:25,520 --> 00:05:26,520 it. 67 00:05:26,520 --> 00:05:31,800 John or anyone else, do you have any thoughts about how much analytics is needed here in 68 00:05:31,800 --> 00:05:32,800 the cannabis space? 69 00:05:32,800 --> 00:05:33,800 I may be buying. 70 00:05:33,800 --> 00:05:36,800 Well, I could go on forever. 71 00:05:36,800 --> 00:05:43,120 But as I said, people are getting more savvy to it. 72 00:05:43,120 --> 00:05:45,120 It's not even just the cannabis space. 73 00:05:45,120 --> 00:05:47,920 It's more just industry in general. 74 00:05:47,920 --> 00:05:59,160 The way I describe it was maybe 2015 to 2018 or so were the data-basic years. 75 00:05:59,160 --> 00:06:02,840 Those were basically where people were just trying to figure out how can we just put a 76 00:06:02,840 --> 00:06:06,800 cap on this fire hose of data coming out. 77 00:06:06,800 --> 00:06:13,560 People were trying to figure out how to collect it, what could be collected, where to store 78 00:06:13,560 --> 00:06:19,800 it, how to store it, what the data should look like, what should standards be, all that 79 00:06:19,800 --> 00:06:20,800 jazz. 80 00:06:20,800 --> 00:06:23,200 And that's largely been figured out. 81 00:06:23,200 --> 00:06:25,920 And so now people have these giant databases. 82 00:06:25,920 --> 00:06:29,080 So now people have all their data. 83 00:06:29,080 --> 00:06:34,600 It may have been cleaned by 2015 standards, but so be it. 84 00:06:34,600 --> 00:06:35,600 We'll take it. 85 00:06:35,600 --> 00:06:41,840 So now they've got their data, and now it's time to do something with it. 86 00:06:41,840 --> 00:06:49,840 And the issue is severalfold. 87 00:06:49,840 --> 00:06:57,280 The data sets in the cannabis space are by and large pretty narrow in focus. 88 00:06:57,280 --> 00:07:03,360 Part of that is historical and as regulatory as come on, they've principally focused on 89 00:07:03,360 --> 00:07:09,320 product safety and product safety attributes that can be measured. 90 00:07:09,320 --> 00:07:14,920 Again, California took the lead because of their pesticide focus. 91 00:07:14,920 --> 00:07:23,440 And it really drove pesticide analysis big time in the analytics space, microbial testing, 92 00:07:23,440 --> 00:07:26,000 and the cannabinoid levels as they come online. 93 00:07:26,000 --> 00:07:27,500 But that's about it. 94 00:07:27,500 --> 00:07:35,160 We tried very hard during when there was public comment period in California during the run 95 00:07:35,160 --> 00:07:38,740 up to the Prop 68. 96 00:07:38,740 --> 00:07:43,320 And we tried to push pretty hard that they should be expanding it beyond just cannabinoids. 97 00:07:43,320 --> 00:07:48,680 And we pushed hard for terpene analysis to be required by regulatory. 98 00:07:48,680 --> 00:07:49,800 And that never happened. 99 00:07:49,800 --> 00:07:53,220 And so that's hit or miss. 100 00:07:53,220 --> 00:08:02,320 The field is moving towards having terpene analysis more prevalent in the commodity. 101 00:08:02,320 --> 00:08:04,780 Now consumers are starting to demand it. 102 00:08:04,780 --> 00:08:07,000 So you're starting to see it. 103 00:08:07,000 --> 00:08:14,080 But the other key compounds that would be affecting the outcome, namely esters and thiles, 104 00:08:14,080 --> 00:08:18,600 sulfur containing compounds, are not really part of the analysis yet. 105 00:08:18,600 --> 00:08:26,400 So we get a very narrow perspective on what you can tease out and interpret. 106 00:08:26,400 --> 00:08:29,120 Anyway, I'll stop there. 107 00:08:29,120 --> 00:08:35,480 But I would like to see broader types of analyses. 108 00:08:35,480 --> 00:08:40,320 Again, it's a marketing question because what's the ROI? 109 00:08:40,320 --> 00:08:46,760 But at the end of the day, I think cannabis, it's pretty narrow in its focus. 110 00:08:46,760 --> 00:08:54,560 And so there's only so much you can get from narrow data sets. 111 00:08:54,560 --> 00:08:57,720 And that's where the cannabis data science team comes in. 112 00:08:57,720 --> 00:09:02,720 And I think that's where sort of the lesson of the day will fit in so nicely because I'll 113 00:09:02,720 --> 00:09:09,440 show you how we've been stuck on these real easy low-hanging fruit. 114 00:09:09,440 --> 00:09:14,280 And I'll show you how we can use a cherry picker to go and get the hard to reach fruit 115 00:09:14,280 --> 00:09:20,760 today because that's what we're all about, using clever tools to get the juicy fruit 116 00:09:20,760 --> 00:09:24,420 that's just out of reach. 117 00:09:24,420 --> 00:09:31,240 So here is one of these certificates of analysis that John is talking about. 118 00:09:31,240 --> 00:09:41,880 So this is, I think, just a mock product or maybe not, but just some flower that was sampled 119 00:09:41,880 --> 00:09:47,080 here in January. 120 00:09:47,080 --> 00:09:52,000 And you can see, okay, it passed the overall tests. 121 00:09:52,000 --> 00:10:00,840 And typically, what people will put on their label is they'll just put maybe the total 122 00:10:00,840 --> 00:10:13,480 THC, the total CBD, and the total cannabinoids because chances are the vendor who purchased 123 00:10:13,480 --> 00:10:20,640 – so at the retailer, right, some vendor purchased this flower from a wholesaler, like 124 00:10:20,640 --> 00:10:24,440 a producer in California, it would be a distributor. 125 00:10:24,440 --> 00:10:27,920 So this distributor sold this product to a retailer. 126 00:10:27,920 --> 00:10:33,440 They're putting it on the shelf, and all they have is this PDF. 127 00:10:33,440 --> 00:10:36,600 So this is cool, right? 128 00:10:36,600 --> 00:10:41,440 So I always say the labs, their product is twofold. 129 00:10:41,440 --> 00:10:49,760 One, it's the certificate saying, yes, this got signed off on by the laboratory director 130 00:10:49,760 --> 00:10:52,320 at Greenleaf Labs. 131 00:10:52,320 --> 00:10:55,960 And then the second part are the data points. 132 00:10:55,960 --> 00:11:03,280 So the retailer could try to copy and paste this and put this into Excel. 133 00:11:03,280 --> 00:11:06,200 And so they probably do that to a certain extent, right? 134 00:11:06,200 --> 00:11:11,960 They probably grab the cannabinoids and they make their labels. 135 00:11:11,960 --> 00:11:16,520 However, I mean, look at all these data points here. 136 00:11:16,520 --> 00:11:18,960 And so this is what John was talking about. 137 00:11:18,960 --> 00:11:24,560 So now you come down and, okay, this sample was tested for terpenes. 138 00:11:24,560 --> 00:11:33,760 So now you have all the monoterpenes, incestuous terpenes, but that's going to, you know, you're 139 00:11:33,760 --> 00:11:37,800 going to have to start paying someone by the hour, right? 140 00:11:37,800 --> 00:11:42,640 If you're just going to have somebody copy and paste this into Excel, you're getting 141 00:11:42,640 --> 00:11:45,600 dozens of these a day, maybe. 142 00:11:45,600 --> 00:11:47,960 Maybe you've got hundreds of these, right? 143 00:11:47,960 --> 00:11:56,360 You're looking at quite a cost if you just want to pay someone to copy and paste this 144 00:11:56,360 --> 00:12:00,760 into Excel. 145 00:12:00,760 --> 00:12:09,480 And then, oh, man, like, you know, now you've got all the pesticides down here. 146 00:12:09,480 --> 00:12:22,640 So all the pesticides, heavy metals, microtoxins, which are produced by microbes. 147 00:12:22,640 --> 00:12:26,480 And so then you also have your product screen. 148 00:12:26,480 --> 00:12:27,480 And double check beyond that. 149 00:12:27,480 --> 00:12:30,080 I'm not a microbiologist. 150 00:12:30,080 --> 00:12:31,400 Definitely double check. 151 00:12:31,400 --> 00:12:33,040 That's a whole field of its own. 152 00:12:33,040 --> 00:12:35,480 That's what's cool about the laboratory, right? 153 00:12:35,480 --> 00:12:44,200 You may have microbiologists working who know these analytes really well. 154 00:12:44,200 --> 00:12:48,520 You may have a chemist who's performing your heavy metal test. 155 00:12:48,520 --> 00:12:53,080 You may have another chemist performing your pesticide test. 156 00:12:53,080 --> 00:12:57,760 So this is awesome data, right? 157 00:12:57,760 --> 00:13:02,720 These are top-notch scientists producing this data. 158 00:13:02,720 --> 00:13:12,720 So why just let this go to the garbage bin, but why just let this collect dust in your 159 00:13:12,720 --> 00:13:13,720 drive? 160 00:13:13,720 --> 00:13:20,280 Was there a question or comment? 161 00:13:20,280 --> 00:13:22,800 Yeah. 162 00:13:22,800 --> 00:13:27,760 So you mentioned that this is to some degree open source. 163 00:13:27,760 --> 00:13:33,800 I guess I'm just curious now, like, how you get access to these types of compliance testing 164 00:13:33,800 --> 00:13:35,080 reports and whatnot. 165 00:13:35,080 --> 00:13:36,520 And I don't know. 166 00:13:36,520 --> 00:13:38,160 I'll probably learn more. 167 00:13:38,160 --> 00:13:43,160 I don't have too much time left until my next meeting, but I'll be coming next week to kind 168 00:13:43,160 --> 00:13:45,480 of a longer meeting to talk more. 169 00:13:45,480 --> 00:13:47,600 Graham, where are you? 170 00:13:47,600 --> 00:13:49,600 I'm in Salt Lake City right now. 171 00:13:49,600 --> 00:13:50,600 Yeah. 172 00:13:50,600 --> 00:13:55,720 So you're going to be more limited in Utah than you are in jurisdictions like Washington, 173 00:13:55,720 --> 00:14:00,320 California, Oregon, in terms of accessing this. 174 00:14:00,320 --> 00:14:02,400 It's more or less open source. 175 00:14:02,400 --> 00:14:08,600 I believe Washington state is doing a much better job at getting this out in open source 176 00:14:08,600 --> 00:14:09,600 mode. 177 00:14:09,600 --> 00:14:14,000 California, it's all required, but it doesn't circulate. 178 00:14:14,000 --> 00:14:20,680 It circulates at the level of the distributor and the retailer, and the consumer can ask 179 00:14:20,680 --> 00:14:23,360 for it and hopefully get it. 180 00:14:23,360 --> 00:14:28,880 But it's not like it's all online and you can just go download it. 181 00:14:28,880 --> 00:14:34,800 And Grant, I'll have to point you to our episode from last week. 182 00:14:34,800 --> 00:14:40,880 I'll try to get that uploaded for you because basically the idea is these tests are being 183 00:14:40,880 --> 00:14:48,360 performed for consumer safety. 184 00:14:48,360 --> 00:14:55,840 So if you're a consumer, you should be able to access your data here. 185 00:14:55,840 --> 00:15:13,920 So let me – for example, let's say one of your products – so here is a product that 186 00:15:13,920 --> 00:15:19,120 I recently purchased here in California. 187 00:15:19,120 --> 00:15:28,240 So you can get this product and then on the label it'll have a – it may have a QR code. 188 00:15:28,240 --> 00:15:31,760 And so here you could find these results online. 189 00:15:31,760 --> 00:15:38,600 And so the idea is at the end of the day the reason the quality assurance is mandated is 190 00:15:38,600 --> 00:15:44,000 to get the consumer educated about their product. 191 00:15:44,000 --> 00:15:46,720 And so that's sort of what Canalytics is here to solve. 192 00:15:46,720 --> 00:15:52,160 We're just saying, okay, they've done all this fantastic quality assurance testing, 193 00:15:52,160 --> 00:16:01,520 but now we need to do the final step and get this data into the hands of the consumer. 194 00:16:01,520 --> 00:16:06,120 So this is a publicly available certificate. 195 00:16:06,120 --> 00:16:11,840 So this is a public URL. 196 00:16:11,840 --> 00:16:18,440 You can browse around this and go download this certificate. 197 00:16:18,440 --> 00:16:23,160 And as a consumer you want to know what's in your product. 198 00:16:23,160 --> 00:16:28,520 You want to know your cannabinoids, your terpenes. 199 00:16:28,520 --> 00:16:38,000 And basically having the PDF is cool, but what we're going to do – and we've already 200 00:16:38,000 --> 00:16:44,840 begun with – we've already checked off confident cannabis, tag-leaf limbs, and then 201 00:16:44,840 --> 00:16:47,760 I'll show you today Greenleaf Labs. 202 00:16:47,760 --> 00:16:52,480 But basically we can just start parsing these COAs. 203 00:16:52,480 --> 00:16:59,040 That way you can actually walk away at the end of the day with – whether it's a JSON 204 00:16:59,040 --> 00:17:04,160 file or just a simple Excel file, you can just walk away at the end of the day with 205 00:17:04,160 --> 00:17:14,000 all of this data in an Excel file that you can give to your favorite data scientist. 206 00:17:14,000 --> 00:17:20,200 You can call Grant up on the phone and say, hey, I've got all of my data now. 207 00:17:20,200 --> 00:17:24,200 Can you crank some numbers on this? 208 00:17:24,200 --> 00:17:29,600 Does that help answer your question? 209 00:17:29,600 --> 00:17:31,600 Yeah, it does. 210 00:17:31,600 --> 00:17:34,320 I really appreciate that. 211 00:17:34,320 --> 00:17:36,320 This is really neat. 212 00:17:36,320 --> 00:17:44,280 So, Gary, I'm not as knowledgeable on terpenes or these other compounds that are in the products, 213 00:17:44,280 --> 00:17:52,640 but I guess I'm just interested now more in what consumers care about and why they 214 00:17:52,640 --> 00:17:53,640 care about it. 215 00:17:53,640 --> 00:17:54,640 Consumers care about getting high. 216 00:17:54,640 --> 00:18:02,620 If you want to know that, go be a bud zender. 217 00:18:02,620 --> 00:18:07,800 Do it for a week and you will learn so much. 218 00:18:07,800 --> 00:18:09,680 That's actually a good recommendation. 219 00:18:09,680 --> 00:18:23,080 In fact, I have heard somebody make the same recommendation that if you truly want to learn 220 00:18:23,080 --> 00:18:29,280 a lot, you spend a week as a bud tender. 221 00:18:29,280 --> 00:18:31,320 Different consumers will want different things. 222 00:18:31,320 --> 00:18:35,600 Some just want the highest cannabinoid concentration. 223 00:18:35,600 --> 00:18:38,760 A lot – I mean, think about it. 224 00:18:38,760 --> 00:18:44,440 Here we're having the conversation about these analytes, but we're probably on the far end 225 00:18:44,440 --> 00:18:50,640 of the distribution of people who are educated about this. 226 00:18:50,640 --> 00:18:56,480 What I have found is the more that people find out about, say, terpenes or, in fact, 227 00:18:56,480 --> 00:18:59,480 pesticides, the more they care about them. 228 00:18:59,480 --> 00:19:06,380 For example, I, in particular, after spending time at the laboratory, I'm particularly 229 00:19:06,380 --> 00:19:13,160 interested in seeing pesticide results because what people don't realize is, yes, you may 230 00:19:13,160 --> 00:19:23,000 have a pesticide pass, but I want to then go and make sure that it's non-detect across 231 00:19:23,000 --> 00:19:25,440 the board. 232 00:19:25,440 --> 00:19:31,160 In this one, it did detect for pyrethrins, but you may have to double-check on that because 233 00:19:31,160 --> 00:19:39,440 it's possible that there may be some naturally occurring pyrethrins. 234 00:19:39,440 --> 00:19:46,600 But now as an educated consumer, you can now go and do your homework. 235 00:19:46,600 --> 00:19:52,480 Now you don't have to research 70-some pesticides. 236 00:19:52,480 --> 00:20:04,120 You can now just go and say, oh, should I be concerned about 0.1 nanogram per gram or 237 00:20:04,120 --> 00:20:06,880 we'll have to find the units here. 238 00:20:06,880 --> 00:20:12,600 Yeah, UG per G. I believe that's nanogram per gram. 239 00:20:12,600 --> 00:20:18,760 Long story short, from my background, I believe there's some naturally occurring pyrethrins, 240 00:20:18,760 --> 00:20:27,120 so I think this one you may not – once again, do your own homework. 241 00:20:27,120 --> 00:20:32,840 I would say do a quick search on pyrethrins and you'll probably learn more than I can 242 00:20:32,840 --> 00:20:34,280 tell you here. 243 00:20:34,280 --> 00:20:42,800 But this is what I would look for as a consumer, is I would look for any detects. 244 00:20:42,800 --> 00:20:47,120 I don't think producers are going to love to hear that news, but like I said, I think 245 00:20:47,120 --> 00:20:56,360 it's just – it'll just kind of make everybody try a little bit harder because – and not 246 00:20:56,360 --> 00:21:03,880 every consumer would care about this, but me in particular, this is what I look for. 247 00:21:03,880 --> 00:21:08,320 And then the second thing I look for are just terpenes, right? 248 00:21:08,320 --> 00:21:16,880 So just again, speaking from personal preference, I've recently kind of learned that limonene 249 00:21:16,880 --> 00:21:23,560 may be what puts me on the couch. 250 00:21:23,560 --> 00:21:34,960 So I maybe would want to try to avoid limonene if I'm wanting to be productive, so to speak. 251 00:21:34,960 --> 00:21:40,080 But once again, this is getting into a whole can of worms and this is where you may want 252 00:21:40,080 --> 00:21:45,000 to speak with John because this is sort of right up John's alley and the work he's 253 00:21:45,000 --> 00:21:56,000 doing is you're consuming many, many milligrams of many different compounds here and what 254 00:21:56,000 --> 00:21:58,440 effect is this having on your body? 255 00:21:58,440 --> 00:22:04,400 Are you kind of moving all these chemical dials in your brain? 256 00:22:04,400 --> 00:22:11,200 But a lot of it's just trial and error and people are just kind of just figuring it out 257 00:22:11,200 --> 00:22:12,200 this haphazardly. 258 00:22:12,200 --> 00:22:19,640 I've heard a saying once that if you're not measuring it, you're not managing it. 259 00:22:19,640 --> 00:22:27,040 And once again, I've basically kind of found that I sort of gravitate towards cannabis 260 00:22:27,040 --> 00:22:34,040 that has terpenelene in it and it took me forever to figure this out. 261 00:22:34,040 --> 00:22:37,680 And then it's like what is a lot of terpenelene? 262 00:22:37,680 --> 00:22:48,880 And we've done the statistics here and I think maybe.2 or.25 is maybe the average around 263 00:22:48,880 --> 00:22:51,120 the mean or median. 264 00:22:51,120 --> 00:22:55,320 So this would be above average terpenelene. 265 00:22:55,320 --> 00:23:00,720 Once again, I'm just pulling that number out of my memory, so go double check us on this. 266 00:23:00,720 --> 00:23:07,080 But that's the thing is we're sort of these real educated consumers and so the idea is 267 00:23:07,080 --> 00:23:13,160 can we get this data to more and more people so that way hopefully other people can have 268 00:23:13,160 --> 00:23:21,800 these revelations about what works well for them and what doesn't because we're also 269 00:23:21,800 --> 00:23:24,840 kind of finding out that everybody's a little different. 270 00:23:24,840 --> 00:23:27,400 We all have our different biochemistries. 271 00:23:27,400 --> 00:23:35,740 So we all have individual biochemistry and then you're mixing in this complex chemistry 272 00:23:35,740 --> 00:23:41,160 and now we kind of need to connect the dots. 273 00:23:41,160 --> 00:23:46,360 So that's one of the reasons Cantalytics is helping with this is we're trying to help 274 00:23:46,360 --> 00:23:54,480 people like John who essentially are doing clinical studies. 275 00:23:54,480 --> 00:24:00,520 John's doing the dosing project where people are consuming cannabis and reporting effects 276 00:24:00,520 --> 00:24:11,240 and now if we can get these data connected to all of that then people can start to figure 277 00:24:11,240 --> 00:24:18,840 out what a reasonable dose may look like and how they could track their consumption over 278 00:24:18,840 --> 00:24:19,840 time. 279 00:24:19,840 --> 00:24:21,840 That's really neat. 280 00:24:21,840 --> 00:24:22,840 I apologize. 281 00:24:22,840 --> 00:24:27,360 I have to step out for a work meeting now but it's been really good to meet everyone 282 00:24:27,360 --> 00:24:30,020 here and you'll see me again. 283 00:24:30,020 --> 00:24:33,680 And Grant thanks for tuning in and tune in anytime. 284 00:24:33,680 --> 00:24:34,680 Yeah appreciate it. 285 00:24:34,680 --> 00:24:36,680 Have a good day everyone. 286 00:24:36,680 --> 00:24:38,840 Thank you. 287 00:24:38,840 --> 00:24:47,720 Well since we can kind of change gears now and since I believe everyone here for the 288 00:24:47,720 --> 00:24:55,320 most part is a you know COA aficionado do you want to see how to start parsing these? 289 00:24:55,320 --> 00:24:59,120 Yeah I'm curious to see how this is working and then I have a question for you based on 290 00:24:59,120 --> 00:25:00,120 that. 291 00:25:00,120 --> 00:25:01,120 Yes please. 292 00:25:01,120 --> 00:25:04,360 So I'll show you how it works. 293 00:25:04,360 --> 00:25:12,520 So essentially the way I've figured the way I like to do a lot of programming is all the 294 00:25:12,520 --> 00:25:20,600 constants that you can sort of was there a question or comment? 295 00:25:20,600 --> 00:25:28,960 So all the constants that you sort of can pull out into a JSON object the better. 296 00:25:28,960 --> 00:25:35,160 So here's sort of just a Greenleaf lab observation. 297 00:25:35,160 --> 00:25:42,080 Give me one second. 298 00:25:42,080 --> 00:25:45,160 I may have to restart this. 299 00:25:45,160 --> 00:25:58,080 Give me one second. 300 00:25:58,080 --> 00:26:05,160 Okay so basically we'll try to pull out all of the constants so that way you know you 301 00:26:05,160 --> 00:26:07,680 could store this in your database. 302 00:26:07,680 --> 00:26:14,720 And you know this would be a typical database entry you know for your given lab right. 303 00:26:14,720 --> 00:26:19,560 You've got Greenleaf lab and you've got all their details about them. 304 00:26:19,560 --> 00:26:27,600 And so then the idea is we'd like to make the analyses more variable than this but in 305 00:26:27,600 --> 00:26:37,960 this case I just went ahead and looked at Greenleaf lab COA and I'm like okay you know 306 00:26:37,960 --> 00:26:42,040 here are their analyses. 307 00:26:42,040 --> 00:26:47,840 And it may seem like a little bit of work but hopefully you should only really have 308 00:26:47,840 --> 00:26:53,520 to do this you know once or twice because you know I visited labs all across the country 309 00:26:53,520 --> 00:27:00,360 and these are basically the standard analyses that you'll have at laboratories right. 310 00:27:00,360 --> 00:27:06,440 Cannabinoids, pesticides, water activity and moisture content. 311 00:27:06,440 --> 00:27:10,720 You'll have terpenes right that may be optional so you should basically view all of these 312 00:27:10,720 --> 00:27:13,720 analyses as optional. 313 00:27:13,720 --> 00:27:20,080 Then you've got heavy metals, lycotoxins, microbial and then foreign matter. 314 00:27:20,080 --> 00:27:24,920 And then in California they call it foreign material so that's a curve ball but you know 315 00:27:24,920 --> 00:27:26,920 we can work with that. 316 00:27:26,920 --> 00:27:35,120 So once again we'd try to avoid that so if there's any way we could sort of figure this 317 00:27:35,120 --> 00:27:40,240 out through natural language processing that would be the best right. 318 00:27:40,240 --> 00:27:46,000 So we could just read microbials by PCR and know that's microbials. 319 00:27:46,000 --> 00:27:52,840 And as you can see you could come close to doing that but I've just predefined them. 320 00:27:52,840 --> 00:27:59,400 Once again natural language processing could probably solve that but that's using quite 321 00:27:59,400 --> 00:28:04,920 the heavy power tool for a small task. 322 00:28:04,920 --> 00:28:10,600 And then basically the way I'm approaching this is I'm going to basically divide the 323 00:28:10,600 --> 00:28:12,280 page up. 324 00:28:12,280 --> 00:28:17,060 So I'll just show you how I'll do that real quick. 325 00:28:17,060 --> 00:28:23,320 So the idea is we can read in this PDF. 326 00:28:23,320 --> 00:28:30,080 One second here. 327 00:28:30,080 --> 00:28:38,080 The idea is we can read in the PDF and here I'll just look at the first page and this 328 00:28:38,080 --> 00:28:44,960 will just be sort of all the rectangles that we can detect on the first page. 329 00:28:44,960 --> 00:28:56,480 So you can see okay awesome like we can detect this area looks like kind of looks like a 330 00:28:56,480 --> 00:29:03,080 table rectangle area so we can get that data real easily. 331 00:29:03,080 --> 00:29:10,000 And so the way I approach this is and this is why I was saying this could almost be a 332 00:29:10,000 --> 00:29:13,160 lab by lab project that you could potentially get paid for right. 333 00:29:13,160 --> 00:29:19,720 So say you go to a retailer and they say hey we've got labs for or we've got COAs from 334 00:29:19,720 --> 00:29:23,960 these 10 labs. 335 00:29:23,960 --> 00:29:30,080 You may want to you know that could be a good contract for you. 336 00:29:30,080 --> 00:29:33,440 But once again it could pay off big time for them right. 337 00:29:33,440 --> 00:29:42,280 If they're able to unlock COAs from 10 different laboratories for you know the foreseeable 338 00:29:42,280 --> 00:29:49,140 future and all their historic COAs and that's a lot of value to be added. 339 00:29:49,140 --> 00:29:53,720 So once again and I'll show you how you can kind of make short work of this and hopefully 340 00:29:53,720 --> 00:29:57,560 it'll get faster and faster each COA we parse. 341 00:29:57,560 --> 00:30:02,760 But the idea is you can just start to essentially use computer vision right. 342 00:30:02,760 --> 00:30:08,680 And that's why I was saying you know that's what's so cool about the PDF plumber package 343 00:30:08,680 --> 00:30:17,320 is it sort of works on you know kind of I believe it leverages sort of you know smart 344 00:30:17,320 --> 00:30:19,440 parsing techniques under the hood. 345 00:30:19,440 --> 00:30:24,080 So if you're interested you can get the images. 346 00:30:24,080 --> 00:30:29,800 So I'll show you down below how you could actually get this image data. 347 00:30:29,800 --> 00:30:37,160 So if you were predisposed you could get that image and display it on your website if you 348 00:30:37,160 --> 00:30:40,440 wanted to. 349 00:30:40,440 --> 00:30:48,840 This is where you kind of get into who owns the copyright of that image and what exactly 350 00:30:48,840 --> 00:30:49,840 is copywritten. 351 00:30:49,840 --> 00:30:58,440 And so this is where I would almost write to like who owns like the raw data of that 352 00:30:58,440 --> 00:30:59,440 image. 353 00:30:59,440 --> 00:31:06,960 I guess right like is the image data itself part of the COA product right. 354 00:31:06,960 --> 00:31:16,920 It's like the image data is similar to like a cannabinoid result or is the image like 355 00:31:16,920 --> 00:31:19,920 a proprietary Greenleaf lab image. 356 00:31:19,920 --> 00:31:23,080 So I don't know. 357 00:31:23,080 --> 00:31:30,360 I suspect it goes with the client. 358 00:31:30,360 --> 00:31:39,040 Exactly so you know if the client paid for the COA and Greenleaf lab took that photo 359 00:31:39,040 --> 00:31:43,240 you know can the client then do anything they want with that photo. 360 00:31:43,240 --> 00:31:46,040 Yeah that's part of the compliance package. 361 00:31:46,040 --> 00:31:49,400 That's what I that would be my argument. 362 00:31:49,400 --> 00:31:53,320 Of course Greenleaf lab owns their own logo but I would say you know if you paid for your 363 00:31:53,320 --> 00:32:00,160 test then I would say your image is part of your result. 364 00:32:00,160 --> 00:32:04,720 But anywho that's getting a little nitpicky but these are the kind of the cool things 365 00:32:04,720 --> 00:32:11,360 that we can start figuring out here in you know this digital 21st century. 366 00:32:11,360 --> 00:32:19,720 Because like I said the image data is quite rich and as we're figuring out you know some 367 00:32:19,720 --> 00:32:24,600 of these image processing techniques can provide a lot of data. 368 00:32:24,600 --> 00:32:29,520 But we're just more interested in the words. 369 00:32:29,520 --> 00:32:36,600 So basically whenever you get a PDF I would say you know the first thing you should do 370 00:32:36,600 --> 00:32:45,160 is just look at all of the words on the page just to make sure that you can get everything. 371 00:32:45,160 --> 00:32:53,440 And so here I'm just looking at the front page but we can look at you know subsequent 372 00:32:53,440 --> 00:32:57,780 pages here. 373 00:32:57,780 --> 00:33:03,760 This one we'll need to make sure that we can in fact get this area here. 374 00:33:03,760 --> 00:33:09,080 Have no fear I think we can do it. 375 00:33:09,080 --> 00:33:14,680 But that's sort of the first thing is you know make sure that you can extract the words. 376 00:33:14,680 --> 00:33:20,680 Have no fear we can get those ones. 377 00:33:20,680 --> 00:33:30,160 But I think we may have to get down to the character level. 378 00:33:30,160 --> 00:33:35,800 So here if you just look at all the characters yes so that's the thing is right couldn't 379 00:33:35,800 --> 00:33:40,920 find all the words but look we can get all of the characters. 380 00:33:40,920 --> 00:33:50,280 So as long as you know we can get every last data point you know there should be a way 381 00:33:50,280 --> 00:33:55,120 to write a clever enough algorithm to actually get the data points. 382 00:33:55,120 --> 00:34:01,480 So that's sort of the first step is just make sure that you have like an actual PDF and 383 00:34:01,480 --> 00:34:08,040 not like a right because you may get wild things right you may get like an image or 384 00:34:08,040 --> 00:34:15,600 something that's you know or some sort of and this would be a good detection right into 385 00:34:15,600 --> 00:34:20,600 this you know not that I would expect them but you know I did talk to someone once in 386 00:34:20,600 --> 00:34:26,720 Oregon who mentioned that this was a problem with like you know counterfeit COAs but this 387 00:34:26,720 --> 00:34:32,400 would you know detect immediately if something was say a counterfeit COA because you just 388 00:34:32,400 --> 00:34:38,800 wouldn't be able to read the characters. 389 00:34:38,800 --> 00:34:45,400 But anywho like I said I don't think that's going to be very common. 390 00:34:45,400 --> 00:34:50,960 But the way I go about parsing this is by the page area. 391 00:34:50,960 --> 00:35:06,920 So basically you know if you look at the front page so if you look at the COA which I recommend 392 00:35:06,920 --> 00:35:12,600 you do first thing is basically you just want to look at the COA and just take stock of 393 00:35:12,600 --> 00:35:19,160 every single data point on the COA and then think about you know how you know we're going 394 00:35:19,160 --> 00:35:24,000 to get every last data point off of this COA. 395 00:35:24,000 --> 00:35:33,120 So basically we've got the distributor, we've got the cultivator, we've got the sample details, 396 00:35:33,120 --> 00:35:40,400 we've got the status, we could potentially get the director and stuff but I'm not doing 397 00:35:40,400 --> 00:35:42,800 that for now. 398 00:35:42,800 --> 00:35:48,520 If we see reason two in the future we could but I don't see a reason. 399 00:35:48,520 --> 00:35:56,600 But then we see okay we've got this information repeated so we only need to get that once 400 00:35:56,600 --> 00:36:03,840 and then it just repeats and then we've got new data here in this table and then we've 401 00:36:03,840 --> 00:36:06,320 got a repeated footer. 402 00:36:06,320 --> 00:36:15,080 And so this is a common pattern that you can expect on a certificate of analysis is a common 403 00:36:15,080 --> 00:36:22,320 header a common footer with your meat in the middle. 404 00:36:22,320 --> 00:36:32,560 So the way I approached this was I was like okay I'll just read the distributor data and 405 00:36:32,560 --> 00:36:44,000 read the producer data so you can get the distributor, get the manufacturer and I was 406 00:36:44,000 --> 00:36:51,800 saying okay you know then you could read the sample details so you could specify you know 407 00:36:51,800 --> 00:36:59,280 these three areas and you know one may expect that different labs may put these in different 408 00:36:59,280 --> 00:37:06,560 areas or potentially Greenleaf labs could update their COA and things may change but 409 00:37:06,560 --> 00:37:10,920 hopefully you could target these areas. 410 00:37:10,920 --> 00:37:15,480 And basically I was actually describing this to John the other day but I'll just tell you 411 00:37:15,480 --> 00:37:22,680 so basically there's two approaches you could take to parsing these COAs. 412 00:37:22,680 --> 00:37:34,560 There's the blunt crude approach which would just be to read everything and then just try 413 00:37:34,560 --> 00:37:42,920 to split the data at known points so we'll just split lab samples, split at sampling 414 00:37:42,920 --> 00:37:44,660 method. 415 00:37:44,660 --> 00:37:51,780 So that was the route I originally went down but then you have to know every single field 416 00:37:51,780 --> 00:37:55,040 on the COA ahead of time. 417 00:37:55,040 --> 00:38:01,760 So if you do know that so say you're the limbs provider for Greenleaf labs then yes that 418 00:38:01,760 --> 00:38:06,520 would be the way you'd want to go but the way that I'm going to do this is basically 419 00:38:06,520 --> 00:38:14,440 just partition the COA into different areas and then just divide and conquer. 420 00:38:14,440 --> 00:38:21,000 So I'll just get the distributor data, get the producer data, get the sample details 421 00:38:21,000 --> 00:38:29,160 and then look you can get the results, just get this block of results. 422 00:38:29,160 --> 00:38:38,560 And so that way if you're trying to get these results and you cut out all of this noise 423 00:38:38,560 --> 00:38:47,060 then it's going to make parsing this much, much easier and I'll show you that right now. 424 00:38:47,060 --> 00:38:54,360 So basically I'll just show you, we've talked the talk, let's walk the walk. 425 00:38:54,360 --> 00:39:03,600 So let's cut out the distributor details from the front page. 426 00:39:03,600 --> 00:39:13,480 So as you can see it takes a little bit of cleaning to, right, so we make our crop and 427 00:39:13,480 --> 00:39:20,480 we just get this, right, we can read this block of text, right. 428 00:39:20,480 --> 00:39:30,620 So that's this distributor shimmons consulting corporation, distributor shimmons consulting 429 00:39:30,620 --> 00:39:32,120 corporation. 430 00:39:32,120 --> 00:39:35,600 So we can read that text in pretty cleanly. 431 00:39:35,600 --> 00:39:45,320 So the idea is it's pretty easy to parse that crop of text and it's fairly easy to parse 432 00:39:45,320 --> 00:39:50,840 the producer crop of text. 433 00:39:50,840 --> 00:40:00,880 So the idea is if you can specify what areas are these different sections then it can make 434 00:40:00,880 --> 00:40:03,400 parsing it super simple. 435 00:40:03,400 --> 00:40:10,920 And potentially I was thinking you could even build a user interface to make it even simpler, 436 00:40:10,920 --> 00:40:11,920 right. 437 00:40:11,920 --> 00:40:18,760 So your user interface would just be to show the PDF to somebody and just have them click 438 00:40:18,760 --> 00:40:26,840 squares and just say oh that's the square for distributor, that's the square for cultivator, 439 00:40:26,840 --> 00:40:28,160 so on and so forth. 440 00:40:28,160 --> 00:40:34,200 But once again it depends on how many of these you have to do. 441 00:40:34,200 --> 00:40:42,160 Once again just since we've got a little bit of extra time I'll show you how you can get 442 00:40:42,160 --> 00:40:44,280 the image data. 443 00:40:44,280 --> 00:40:58,000 So the COA doc, which I've initialized right here, has a nice built-in function to get 444 00:40:58,000 --> 00:41:00,400 the PDF image data. 445 00:41:00,400 --> 00:41:06,000 So here I want to go ahead and just get the image. 446 00:41:06,000 --> 00:41:13,980 And so this will be the raw data for that image, right. 447 00:41:13,980 --> 00:41:23,800 So this image right here, as far as the computer is concerned, that image may as well be this 448 00:41:23,800 --> 00:41:26,440 block of text. 449 00:41:26,440 --> 00:41:30,960 So this doesn't mean anything to a human. 450 00:41:30,960 --> 00:41:39,360 This is in fact just 34,000 characters of essentially randomness. 451 00:41:39,360 --> 00:41:41,440 But that's your image. 452 00:41:41,440 --> 00:41:48,420 And so the idea is you can give this block of text to your developer and they can display 453 00:41:48,420 --> 00:41:58,880 that on your website or print it on a label if you were so inclined on doing so. 454 00:41:58,880 --> 00:42:05,400 But that's a lot of text so I'm not going to be collecting all that data. 455 00:42:05,400 --> 00:42:11,960 But if you want it, it stays for the taking. 456 00:42:11,960 --> 00:42:17,200 And you can get the sample details. 457 00:42:17,200 --> 00:42:23,040 Once again, the Raspberry Parfait. 458 00:42:23,040 --> 00:42:26,320 So here you see Raspberry Parfait. 459 00:42:26,320 --> 00:42:31,360 And it's a little bit trickier to parse these. 460 00:42:31,360 --> 00:42:38,240 So this block of code is a little bit specialized. 461 00:42:38,240 --> 00:42:46,640 But the idea is if myself and all of you awesome data scientists, if we keep tinkering on this, 462 00:42:46,640 --> 00:42:55,800 hopefully we can generalize these algorithms so that way you can just point at the area 463 00:42:55,800 --> 00:43:05,400 on the PDF that you want parsed and you'll get your data here. 464 00:43:05,400 --> 00:43:12,520 And I'm just going to clear this out, just to save some room here. 465 00:43:12,520 --> 00:43:20,360 So now we've got all the distributor data, we've got all the producer data, we've got 466 00:43:20,360 --> 00:43:22,640 all the sample details. 467 00:43:22,640 --> 00:43:27,360 So that's pretty cool. 468 00:43:27,360 --> 00:43:29,480 We've got the product name, right? 469 00:43:29,480 --> 00:43:30,480 We may not yet. 470 00:43:30,480 --> 00:43:37,320 I was hoping we did. 471 00:43:37,320 --> 00:43:40,160 Well we may not have the product name yet. 472 00:43:40,160 --> 00:43:46,040 Oh, that's right, I added it at the end. 473 00:43:46,040 --> 00:43:48,000 So there it is. 474 00:43:48,000 --> 00:43:49,000 Cool. 475 00:43:49,000 --> 00:43:55,140 So now the idea is divide and conquer. 476 00:43:55,140 --> 00:43:58,280 So basically you need to get a couple things at this point. 477 00:43:58,280 --> 00:44:03,960 And so this is why, you know, CanLytics is here to help you standardize these. 478 00:44:03,960 --> 00:44:11,680 So basically we'll standardize the analyses and so basically at the top I've got, I put 479 00:44:11,680 --> 00:44:15,520 a list of all the data points that we'll collect. 480 00:44:15,520 --> 00:44:23,680 So we basically collect the analyses, collect all the distributor details, and then the 481 00:44:23,680 --> 00:44:26,760 big thing are the results, right? 482 00:44:26,760 --> 00:44:33,800 So we'll want to go ahead and get the analyses real quick just to be complete, but then we'll 483 00:44:33,800 --> 00:44:35,400 get to the results. 484 00:44:35,400 --> 00:44:46,480 We've got all the analyses, product underwent, and then we have the status for the various 485 00:44:46,480 --> 00:44:48,200 tests. 486 00:44:48,200 --> 00:44:52,600 We can get the methods. 487 00:44:52,600 --> 00:45:00,800 As we found out with PSI Labs, the method to the madness, just kidding, the method does 488 00:45:00,800 --> 00:45:04,000 matter, so it's worth keeping. 489 00:45:04,000 --> 00:45:11,720 I would like to tie these methods to the specific analysis, but just for time's sake, I haven't 490 00:45:11,720 --> 00:45:13,240 gotten to that yet. 491 00:45:13,240 --> 00:45:17,160 But we've got the methods just in case. 492 00:45:17,160 --> 00:45:18,880 Now the fun part. 493 00:45:18,880 --> 00:45:30,200 And so now because we've nicely cropped each page, remember for the results, we'll be cropping 494 00:45:30,200 --> 00:45:34,640 the page to the meat. 495 00:45:34,640 --> 00:45:45,280 We can just pull out all those rows, and then we basically have all of that data. 496 00:45:45,280 --> 00:45:55,200 Yes, we have to now organize it with a pretty clever algorithm, but basically we have all 497 00:45:55,200 --> 00:45:56,200 the data, right? 498 00:45:56,200 --> 00:45:59,280 We've got all the rows, right? 499 00:45:59,280 --> 00:46:06,400 We've got all the cannabinoids, and we've got all the terpenes. 500 00:46:06,400 --> 00:46:17,520 And so the way I approach this is in this case, I predefined the columns, but ideally 501 00:46:17,520 --> 00:46:22,240 your algorithm would be smart enough to detect the columns. 502 00:46:22,240 --> 00:46:30,480 But I just went ahead and as you can see up top, I just predefined, okay, your cannabinoids 503 00:46:30,480 --> 00:46:38,360 will have the name, the LOD, LOQ value, milligrams per gram. 504 00:46:38,360 --> 00:46:48,200 On the COA, the name, LOD, LOQ percentage, milligrams per gram. 505 00:46:48,200 --> 00:46:54,680 So it's not perfect, but at least we kind of know what order the data is in. 506 00:46:54,680 --> 00:47:03,840 And as long as Greenleaf Labs isn't changing their COA willy-nilly, and I'll tell you now 507 00:47:03,840 --> 00:47:08,160 that labs don't change their COAs willy-nilly. 508 00:47:08,160 --> 00:47:10,760 They may change them once in a while. 509 00:47:10,760 --> 00:47:19,000 So once in a while the routine may need to be updated, but these are controlled documents 510 00:47:19,000 --> 00:47:26,120 at the laboratory that have to go through a controlled process in order to be changed. 511 00:47:26,120 --> 00:47:36,320 So it kind of puts the brakes on things just changing rapidly overnight. 512 00:47:36,320 --> 00:47:42,760 So as I said, it's not going to be impossible for them to change their COA, but if they 513 00:47:42,760 --> 00:47:52,240 do so we can make a quick pull request to the algorithm and hopefully stay up to speed. 514 00:47:52,240 --> 00:48:00,400 So it's not perfect, but as I said, it's unlocking such rich data that I think the code was worth 515 00:48:00,400 --> 00:48:01,400 writing. 516 00:48:01,400 --> 00:48:10,040 Kieran, what is it doing for you have these, I don't know if I'd call them header and footers 517 00:48:10,040 --> 00:48:19,840 in the tables, but you've got columns with data and then it's interrupted with bars there. 518 00:48:19,840 --> 00:48:24,240 In this case it's black text bars. 519 00:48:24,240 --> 00:48:25,240 This? 520 00:48:25,240 --> 00:48:26,240 Yeah. 521 00:48:26,240 --> 00:48:29,520 So I'm skipping those. 522 00:48:29,520 --> 00:48:34,560 So basically I'm just going through and I'm saying, oh, if it starts with an analysis 523 00:48:34,560 --> 00:48:42,880 name, if it starts with date, time, I'll just skip it. 524 00:48:42,880 --> 00:48:47,560 And so that's here, I'll show you where in the code I'm doing that. 525 00:48:47,560 --> 00:48:50,720 Okay, I see what you're doing now. 526 00:48:50,720 --> 00:48:51,720 Okay. 527 00:48:51,720 --> 00:48:53,680 Well, I forget where in the code that is. 528 00:48:53,680 --> 00:48:57,160 Oh, I'll never see it in the million years. 529 00:48:57,160 --> 00:49:04,880 So that's what's happening is there is still some custom algorithms. 530 00:49:04,880 --> 00:49:13,800 The idea is, and this is the way I approach programming in general is I don't even mind 531 00:49:13,800 --> 00:49:16,380 writing a specific use case. 532 00:49:16,380 --> 00:49:21,400 So I wrote this specific use case and basically I was like, I'm going to parse all the data 533 00:49:21,400 --> 00:49:26,560 out of this COA one way or the other. 534 00:49:26,560 --> 00:49:32,760 Once I found out that yes, we can read every single piece of data on here, that was the 535 00:49:32,760 --> 00:49:33,760 goal. 536 00:49:33,760 --> 00:49:37,800 It was like one way or the other, we're going to read every single piece of data. 537 00:49:37,800 --> 00:49:43,880 And essentially we did exactly that. 538 00:49:43,880 --> 00:49:49,040 So here I'll just show you this right now. 539 00:49:49,040 --> 00:49:58,960 But basically, it also read in low on battery so if I cut out it's because I lost battery. 540 00:49:58,960 --> 00:50:08,400 But basically we've read every single data point on this COA. 541 00:50:08,400 --> 00:50:16,080 So this would be 59 or I think it's 58 data points. 542 00:50:16,080 --> 00:50:28,880 So tons of data and I've tried to be as specific as possible. 543 00:50:28,880 --> 00:50:34,320 I mean not as specific, I tried to be as general as possible so that way you could generalize 544 00:50:34,320 --> 00:50:37,840 the code to more and more labs. 545 00:50:37,840 --> 00:50:45,520 But inevitably I had to write Greenleaf Labs specific code. 546 00:50:45,520 --> 00:50:53,600 So basically I tried to put as many parameters as I could into JSON. 547 00:50:53,600 --> 00:50:56,960 Then I tried to generalize the code as much as possible. 548 00:50:56,960 --> 00:51:08,360 And at the end of the day, I still had to have a Greenleaf Labs specific function. 549 00:51:08,360 --> 00:51:12,880 So here I wrote a parse Greenleaf Labs function. 550 00:51:12,880 --> 00:51:21,240 And so the idea is all of these different functions, the parse Greenleaf Labs function, 551 00:51:21,240 --> 00:51:32,680 the parse Confident Cannabis function, these can be imported pretty quickly into COA doc. 552 00:51:32,680 --> 00:51:38,760 And so the idea is it's sort of this pluggable user interface. 553 00:51:38,760 --> 00:51:45,680 So you can basically just import your various labs. 554 00:51:45,680 --> 00:51:54,120 So here I've done Confident Cannabis, Tag-Leaf Lens, Greenleaf Labs, Data Scientific, and 555 00:51:54,120 --> 00:51:57,120 then SC Labs. 556 00:51:57,120 --> 00:51:58,920 I'm still finishing that one. 557 00:51:58,920 --> 00:52:02,400 And then MCR Labs, we've done the URL. 558 00:52:02,400 --> 00:52:07,320 I don't actually have a COA PDF from them. 559 00:52:07,320 --> 00:52:11,600 But if they put a QR code on their PDF, then we're good to go. 560 00:52:11,600 --> 00:52:25,640 So Keegan, given what you've got so far, can you issue a report that would show how homogeneous 561 00:52:25,640 --> 00:52:29,200 or not the analyte names are? 562 00:52:29,200 --> 00:52:31,720 Because these are all California COAs. 563 00:52:31,720 --> 00:52:38,640 And they should be, for the most part, all synonymous. 564 00:52:38,640 --> 00:52:43,920 But it would be interesting to see how bad that variance is on a California set right 565 00:52:43,920 --> 00:52:46,640 now. 566 00:52:46,640 --> 00:52:48,360 That would be a fun exercise. 567 00:52:48,360 --> 00:52:52,560 How easy is it to do with what you've got right now? 568 00:52:52,560 --> 00:52:59,160 Well, in fact, this was a test that I put Candace on with, I think Candace may have done it 569 00:52:59,160 --> 00:53:03,520 with the PSI Labs data, but we could do it with the SC Labs data. 570 00:53:03,520 --> 00:53:05,240 But we can do it. 571 00:53:05,240 --> 00:53:08,640 And essentially, we can get all the... 572 00:53:08,640 --> 00:53:12,440 I wouldn't get down with the analyte names. 573 00:53:12,440 --> 00:53:20,600 I would at first ask it at the top level of the assay classes themselves. 574 00:53:20,600 --> 00:53:21,600 The analysis. 575 00:53:21,600 --> 00:53:24,080 The analysis classes themselves. 576 00:53:24,080 --> 00:53:26,360 Start 50,000 feet. 577 00:53:26,360 --> 00:53:36,160 And let's just see how good or bad synonymous is in a set of California COAs. 578 00:53:36,160 --> 00:53:43,360 Because that's going to tell us how we have to run our synonym databases. 579 00:53:43,360 --> 00:53:46,200 I think they're actually the analyses are pretty similar. 580 00:53:46,200 --> 00:53:48,360 They could be. 581 00:53:48,360 --> 00:53:56,800 And I would recommend since VEDA is the most recent and is the most closest to current 582 00:53:56,800 --> 00:54:02,960 DCC because they've been now recently licensed. 583 00:54:02,960 --> 00:54:09,280 And what we just got back from VEDA is their current compliance COA. 584 00:54:09,280 --> 00:54:14,960 It might be pretty damn current. 585 00:54:14,960 --> 00:54:18,960 So here... 586 00:54:18,960 --> 00:54:23,240 So long story short is I think the analyses names are similar. 587 00:54:23,240 --> 00:54:28,360 Sometimes people use potency instead of cannabinoids. 588 00:54:28,360 --> 00:54:34,080 For whatever reason, people just like the term potency in the industry. 589 00:54:34,080 --> 00:54:37,880 I just always felt it was just less formal. 590 00:54:37,880 --> 00:54:38,920 It's wrong. 591 00:54:38,920 --> 00:54:42,680 It's totally fucking wrong. 592 00:54:42,680 --> 00:54:44,720 People just prefer it. 593 00:54:44,720 --> 00:54:47,240 It's a hard habit to break. 594 00:54:47,240 --> 00:54:51,000 It's just not right. 595 00:54:51,000 --> 00:54:55,720 The word potency and potent does have its time and place. 596 00:54:55,720 --> 00:55:00,680 But I feel like the word cannabinoids is more appropriate. 597 00:55:00,680 --> 00:55:03,280 It is totally more appropriate. 598 00:55:03,280 --> 00:55:08,080 You cannot define potency without an activity assay. 599 00:55:08,080 --> 00:55:12,680 And that's the last I'm going to be saying this. 600 00:55:12,680 --> 00:55:16,120 But it's nomenclature that you'll see. 601 00:55:16,120 --> 00:55:17,120 And so... 602 00:55:17,120 --> 00:55:19,640 Fucked up cannabis industry bullshit. 603 00:55:19,640 --> 00:55:23,920 And in fact, I even have a way to deal with this. 604 00:55:23,920 --> 00:55:34,240 And the way we do it is we basically just have replacements. 605 00:55:34,240 --> 00:55:39,800 Here basically just common replacements that I'll make on the COA. 606 00:55:39,800 --> 00:55:45,000 But at least we're not bad at the level of the analyses. 607 00:55:45,000 --> 00:55:50,240 Our terms are pretty synonymous, I guess, in the set you've got right now. 608 00:55:50,240 --> 00:55:51,240 Exactly. 609 00:55:51,240 --> 00:55:57,920 And so basically, if there's any oddball case like that, you can basically just define a 610 00:55:57,920 --> 00:56:07,440 new replacement and say, okay, we just want to replace potency with cannabinoids. 611 00:56:07,440 --> 00:56:17,040 So the idea is to make all of these little adjustments as sort of configurable as possible. 612 00:56:17,040 --> 00:56:24,440 And then the idea is you can just configure away the intricacies of everybody's COA. 613 00:56:24,440 --> 00:56:28,160 So say Greenleaf Lab uses the word potency. 614 00:56:28,160 --> 00:56:30,600 That's not going to break anything. 615 00:56:30,600 --> 00:56:34,760 We'll just make that replacement. 616 00:56:34,760 --> 00:56:39,240 Can I make a recommendation? 617 00:56:39,240 --> 00:56:41,080 What you're doing is wonderful. 618 00:56:41,080 --> 00:56:48,440 Do not go outside the flower box, as it were, because it's going to get a lot worse when 619 00:56:48,440 --> 00:56:54,460 you get processed cannabis products and it's going to be confounding. 620 00:56:54,460 --> 00:57:02,480 So I would at this point just be working with flower because it'll become rabbit holes in 621 00:57:02,480 --> 00:57:08,360 a mess once you go off that. 622 00:57:08,360 --> 00:57:11,680 It could get complex once you get to edibles. 623 00:57:11,680 --> 00:57:13,400 Yeah, it's going to be a mess. 624 00:57:13,400 --> 00:57:15,200 We'll deal with that later. 625 00:57:15,200 --> 00:57:19,160 Let's just do flowers. 626 00:57:19,160 --> 00:57:22,360 And so, exactly. 627 00:57:22,360 --> 00:57:29,080 So edibles often are measured in milligrams per gram and then the milligrams per serving 628 00:57:29,080 --> 00:57:30,480 matters. 629 00:57:30,480 --> 00:57:35,240 So the product classes and everything else gets real gnarly. 630 00:57:35,240 --> 00:57:37,440 So edibles could be a whole can of worms. 631 00:57:37,440 --> 00:57:42,320 If anyone ever comes across an edible COA that they want parsed, we could take a look at 632 00:57:42,320 --> 00:57:43,320 it. 633 00:57:43,320 --> 00:57:45,160 But as you said, we may have to take it from the top. 634 00:57:45,160 --> 00:57:48,320 Later, later. 635 00:57:48,320 --> 00:57:51,400 But for now, and I'll let you all get out of here. 636 00:57:51,400 --> 00:57:53,000 I think my battery's about to go. 637 00:57:53,000 --> 00:57:56,720 But for now, as I said, we can get all the results. 638 00:57:56,720 --> 00:58:01,600 It still needs to be flattened and normalized a bit. 639 00:58:01,600 --> 00:58:06,680 I forgot to add the units, so I need to go back and add the units. 640 00:58:06,680 --> 00:58:13,840 But at the moment, you can parse the COA and get all of this awesome data. 641 00:58:13,840 --> 00:58:16,680 And there's still more work to be done. 642 00:58:16,680 --> 00:58:21,880 But I was reading a pretty inspirational blog post. 643 00:58:21,880 --> 00:58:26,920 And it was basically, or actually, I think it was even a Stack Overflow answer, but basically 644 00:58:26,920 --> 00:58:34,560 somebody said that it's sort of miraculous what you can pull off sort of in a business 645 00:58:34,560 --> 00:58:42,840 model sense with good PDF parsing. 646 00:58:42,840 --> 00:58:43,840 It's a tough task. 647 00:58:43,840 --> 00:58:46,480 And not everybody does this well. 648 00:58:46,480 --> 00:58:50,840 And it's just like it's all about attention to detail. 649 00:58:50,840 --> 00:58:58,200 You could do this quick and dirty, but we had meticulous attention to detail to the 650 00:58:58,200 --> 00:59:04,840 point where we can get the sample image, we can get every last piece of data off of this 651 00:59:04,840 --> 00:59:05,840 COA. 652 00:59:05,840 --> 00:59:10,960 And that's what really makes it at the end of the day is the fact that we can get it 653 00:59:10,960 --> 00:59:22,080 all and organize it in a way that you can save it and your favorite data scientist can 654 00:59:22,080 --> 00:59:24,040 use this. 655 00:59:24,040 --> 00:59:27,320 So that's what really makes this special, I think. 656 00:59:27,320 --> 00:59:34,680 And so for any of you that want to work on this, as I said, we've got a pretty well-defined 657 00:59:34,680 --> 00:59:39,080 way to go about collecting this data. 658 00:59:39,080 --> 00:59:46,680 At least for California, we've got a nice list here of about 50 to 60 data points that 659 00:59:46,680 --> 00:59:49,920 you can expect on almost any COA. 660 00:59:49,920 --> 00:59:58,200 And you should be able to readily parse these data points without too much trouble. 661 00:59:58,200 --> 01:00:04,320 And then write some routines and hopefully get paid for your work. 662 01:00:04,320 --> 01:00:07,320 And that's the limit. 663 01:00:07,320 --> 01:00:11,700 And as I said, I think each bit that you tinker on this, I think, will be helping the cannabis 664 01:00:11,700 --> 01:00:15,560 industry out. 665 01:00:15,560 --> 01:00:19,080 So that's the presentation for today. 666 01:00:19,080 --> 01:00:24,960 Any thoughts, comments, questions before we get out of here and enjoy the bright sunny 667 01:00:24,960 --> 01:00:25,960 day? 668 01:00:25,960 --> 01:00:32,200 That's looking good. 669 01:00:32,200 --> 01:00:34,280 And I want to thank you all for coming along. 670 01:00:34,280 --> 01:00:38,600 And your eyes and your ears are helping. 671 01:00:38,600 --> 01:00:39,600 Thanks Jerry. 672 01:00:39,600 --> 01:00:40,600 Thanks Candice. 673 01:00:40,600 --> 01:00:41,600 John. 674 01:00:41,600 --> 01:00:42,600 All right. 675 01:00:42,600 --> 01:00:44,480 Keegan, hopefully we get more. 676 01:00:44,480 --> 01:00:50,920 As I said, we're waiting on Cafe Floor to see what they can roust up for us. 677 01:00:50,920 --> 01:01:00,440 So maybe it gets faster and easier as if you've gotten this far, maybe it's not so hard to 678 01:01:00,440 --> 01:01:04,440 put a new one in and see how fast it parses. 679 01:01:04,440 --> 01:01:05,440 Exactly. 680 01:01:05,440 --> 01:01:10,800 And that's what I was thinking is it's almost the classic learning by doing. 681 01:01:10,800 --> 01:01:38,200 So the more of these that we do, the...