WEBVTT 00:00:09.029 --> 00:00:11.490 Hey everyone, welcome back to Data Nihai with 00:00:11.490 --> 00:00:14.289 Mukundan. I'm your host Mukundan Sankar and today 00:00:14.289 --> 00:00:17.289 we're going to talk deep about a topic that's 00:00:17.289 --> 00:00:19.390 very close to my heart. Something that could 00:00:19.390 --> 00:00:23.769 completely change how we share our work as data 00:00:23.769 --> 00:00:26.429 analysts, data scientists or machine learning 00:00:26.429 --> 00:00:31.230 professionals. So we're going to be talking about 00:00:31.230 --> 00:00:36.210 automating the creation of engaging animated 00:00:37.200 --> 00:00:41.200 And even beautifully designed presentations that 00:00:41.200 --> 00:00:44.979 are straight from your Jupyter notebooks. And 00:00:44.979 --> 00:00:51.380 we use AI for this, right? This is not just about 00:00:51.380 --> 00:00:55.280 saving time. This is about telling better stories 00:00:55.280 --> 00:01:00.420 with your data. And more importantly, this is 00:01:00.420 --> 00:01:05.719 about making your work shine in interviews. And 00:01:05.719 --> 00:01:10.400 even in team reviews, it could be even if you 00:01:10.400 --> 00:01:15.219 want to do a public talk. So let's just dive 00:01:15.219 --> 00:01:20.180 right in. So here's the problem that we all faced 00:01:20.180 --> 00:01:22.319 early on. And this is something I've personally 00:01:22.319 --> 00:01:26.359 faced. I'm hoping if this is something that you 00:01:26.359 --> 00:01:28.519 faced as well, then this is something that can 00:01:28.519 --> 00:01:32.599 work for you. So I want to take you back to... 00:01:32.969 --> 00:01:35.209 when i was just starting out in the data science 00:01:35.209 --> 00:01:40.310 and analytics field like many of you i was building 00:01:40.310 --> 00:01:43.310 all kinds of projects in jupyter notebooks you 00:01:43.310 --> 00:01:45.530 know when you first started out you're probably 00:01:45.530 --> 00:01:48.310 thinking like what kind of projects would stick 00:01:48.310 --> 00:01:55.689 what would get you hired so so i was doing as 00:01:55.689 --> 00:01:58.829 most of my peers were and i'm sure maybe you 00:01:58.829 --> 00:02:02.200 probably you know relate to this trying to see 00:02:02.200 --> 00:02:04.900 what kind of projects maybe on Kaggle .com or 00:02:04.900 --> 00:02:07.299 could be some, you know, government websites. 00:02:07.859 --> 00:02:09.780 You're trying to find data sets to work with. 00:02:10.319 --> 00:02:14.039 And then you analyze this data set in Jupyter 00:02:14.039 --> 00:02:18.219 Notebooks, right? And it could be something like 00:02:18.219 --> 00:02:20.500 a customer churn model that you're trying to 00:02:20.500 --> 00:02:23.560 build or time series forecasting. So these were 00:02:23.560 --> 00:02:24.979 some of the models that I was trying to build. 00:02:26.219 --> 00:02:31.000 So I had the code, the insights. even the charts 00:02:31.000 --> 00:02:37.000 right even the uh you know graphs that uh all 00:02:37.000 --> 00:02:40.280 of those were ready but something still was not 00:02:40.280 --> 00:02:44.319 um you know adding up right something was still 00:02:44.319 --> 00:02:48.360 missing and when it came time to share that work 00:02:48.360 --> 00:02:52.919 whether in an interview or with a stakeholder 00:02:52.919 --> 00:02:59.530 i struggled because the notebook that I had built, 00:02:59.669 --> 00:03:04.629 it wasn't just built for storytelling. It was 00:03:04.629 --> 00:03:06.689 great for exploration, you know, the exploratory 00:03:06.689 --> 00:03:09.590 analysis, great for iteration, all the machine 00:03:09.590 --> 00:03:13.810 learning tasks, data visualization. But for someone 00:03:13.810 --> 00:03:17.889 unfamiliar with the analysis, well, it was overwhelming, 00:03:18.250 --> 00:03:21.909 right? All they saw was just code, maybe a chart 00:03:21.909 --> 00:03:24.830 or two would, if they had scrolled enough. But 00:03:24.830 --> 00:03:30.750 there was no flow as such. no explanation and 00:03:30.750 --> 00:03:39.650 no guidance but so I started creating these slides 00:03:39.650 --> 00:03:49.610 manually copying markdown exporting charts rewriting 00:03:49.610 --> 00:03:56.879 speaker notes and it was just a nightmare I'm 00:03:56.879 --> 00:03:59.879 sure you know that feeling, especially if you've 00:03:59.879 --> 00:04:03.259 worked in this. That's when I thought, what if 00:04:03.259 --> 00:04:09.939 I could just upload my notebook and get some 00:04:09.939 --> 00:04:15.560 well -structured presentation with slides, with 00:04:15.560 --> 00:04:20.839 speaker notes and visuals, well, instantly, right? 00:04:24.459 --> 00:04:28.800 So that brings me to the vision of my app. So 00:04:28.800 --> 00:04:31.100 that's the idea behind this app that I built 00:04:31.100 --> 00:04:34.339 here. It's a Streamlit app. Again, Streamlit 00:04:34.339 --> 00:04:36.839 is a Python -based library, something that I 00:04:36.839 --> 00:04:40.139 use time and again for my projects. And you can 00:04:40.139 --> 00:04:43.240 view some of my other episodes where some of 00:04:43.240 --> 00:04:45.360 my recent episodes, actually, they all talk about 00:04:45.360 --> 00:04:48.300 me using Streamlit library. It's like a front 00:04:48.300 --> 00:04:53.000 -end Python -based library. if you don't know 00:04:53.000 --> 00:04:56.379 how to use javascript which i don't so streamlit 00:04:56.379 --> 00:04:58.759 is like python based so that makes it very easy 00:04:58.759 --> 00:05:01.759 to use right so it's a streamlit app where you 00:05:01.759 --> 00:05:04.939 upload a jupyter notebook and it does all the 00:05:04.939 --> 00:05:10.040 heavy lifting for you and here's what it gives 00:05:10.040 --> 00:05:15.779 you it gives you like a powerpoint file with 00:05:15.779 --> 00:05:23.300 slides and each slide has a clean title ai generated 00:05:23.300 --> 00:05:27.439 content as well and it gives you bullet points 00:05:27.439 --> 00:05:35.040 so um the main aim is also to give you some speaker 00:05:35.040 --> 00:05:39.339 notes which can guide the way you speak to guide 00:05:39.339 --> 00:05:43.579 your narration right and now now the big one 00:05:43.579 --> 00:05:46.860 it gives you ai generated visual diagrams like 00:05:46.860 --> 00:05:50.379 flow charts process trees and conceptual graphics 00:05:50.379 --> 00:05:54.759 so Well, instead of opening PowerPoint and starting 00:05:54.759 --> 00:06:00.079 from a blank canvas, you now get like 70 to 80 00:06:00.079 --> 00:06:03.220 % of your slide deck already done. All from just 00:06:03.220 --> 00:06:05.920 your notebook, right? You still tweak it, of 00:06:05.920 --> 00:06:09.779 course, and add your own polish. But the structure 00:06:09.779 --> 00:06:13.060 is there. The story is there. The visuals are 00:06:13.060 --> 00:06:17.079 there. And sometimes it may not even give you 00:06:17.079 --> 00:06:19.379 visuals, but I think the idea is at least it'll 00:06:19.379 --> 00:06:22.649 give you ideas to generate those visuals. and 00:06:22.649 --> 00:06:24.910 that's just as important because you can create 00:06:24.910 --> 00:06:26.709 those visuals but you don't know what exactly 00:06:26.709 --> 00:06:30.089 to create um and that's where you can save a 00:06:30.089 --> 00:06:33.170 lot of time by just knowing what to create right 00:06:33.170 --> 00:06:37.410 um and you're just like replicating uh what what's 00:06:37.410 --> 00:06:42.290 being asked for you right um so let's break down 00:06:42.290 --> 00:06:45.389 the tech in this app so let's talk about what's 00:06:45.389 --> 00:06:51.379 under the hood first uh we use nb format library 00:06:51.379 --> 00:06:56.959 to parse the uploaded notebook we extract the 00:06:56.959 --> 00:07:00.660 markdown cells and look for any charts that were 00:07:00.660 --> 00:07:05.240 generated using matplotlib or plotly then we 00:07:05.240 --> 00:07:09.240 feed those markdown cells into gpt4 and here's 00:07:09.240 --> 00:07:13.139 where the magic starts gpt then gives us a slight 00:07:13.139 --> 00:07:17.259 title two to three bullet points for that slide 00:07:18.000 --> 00:07:21.579 and speaker notes and oh and of course this is 00:07:21.579 --> 00:07:23.879 the cool part a textual description of the diagram 00:07:23.879 --> 00:07:27.100 so what i mentioned was you know there could 00:07:27.100 --> 00:07:28.779 be times when it gives the diagram but it also 00:07:28.779 --> 00:07:31.779 can give you a textual description of what you 00:07:31.779 --> 00:07:37.259 what you wanted to create right um so then we 00:07:37.259 --> 00:07:39.480 take that diagram description and we feed it 00:07:39.480 --> 00:07:43.120 to a graphvis library or another visual rendering 00:07:43.120 --> 00:07:48.259 library to actually create that diagram So you 00:07:48.259 --> 00:07:51.740 have this textual description now, but you also 00:07:51.740 --> 00:07:54.319 want to create it. And that's when you can use 00:07:54.319 --> 00:07:58.399 graph -based library or any other visual rendering 00:07:58.399 --> 00:08:02.160 library, like I mentioned, to actually go ahead 00:08:02.160 --> 00:08:04.579 and create the diagram and not just suggest it, 00:08:04.620 --> 00:08:07.040 right? Suggesting, yeah, everybody has an opinion. 00:08:08.500 --> 00:08:13.579 What I think is really helpful here is that you 00:08:13.579 --> 00:08:18.540 are getting... ideas and you are also able to 00:08:18.540 --> 00:08:23.100 create it using these libraries so yeah and of 00:08:23.100 --> 00:08:25.920 course it can handle flowcharts decision trees 00:08:25.920 --> 00:08:31.480 linear pipelines and i'm experimenting with more 00:08:31.480 --> 00:08:35.039 smart art style visuals you know what smart art 00:08:35.039 --> 00:08:38.539 is right it's the one in powerpoint you get these 00:08:38.539 --> 00:08:44.759 amazing ideas to you know recreate to create 00:08:44.759 --> 00:08:48.100 diagrams from whatever text you have so something 00:08:48.100 --> 00:08:52.580 like give you process flow diagrams idea how 00:08:52.580 --> 00:08:55.320 to use a process flow in powerpoint so that's 00:08:55.320 --> 00:08:57.200 something you can explore as well but this is 00:08:57.200 --> 00:09:02.720 something that i'm exploring to use as a next 00:09:02.720 --> 00:09:09.279 step in this app so then also use python pptx 00:09:09.279 --> 00:09:13.940 library to stitch everything together into a 00:09:13.940 --> 00:09:18.500 dot pptx presentation so python pptx like python 00:09:18.500 --> 00:09:25.940 hyphen pptx it helps to convert to a pptx presentation 00:09:25.940 --> 00:09:32.039 right with slides content images and speaker 00:09:32.039 --> 00:09:38.519 notes and it's all done in python now why does 00:09:38.519 --> 00:09:42.149 this matter you might be thinking okay Cool. 00:09:42.470 --> 00:09:45.690 Cool project, right? But what does this actually 00:09:45.690 --> 00:09:51.409 change? Here's the thing. Presentation is leverage. 00:09:53.049 --> 00:09:57.269 You might have the best model, the cleanest data 00:09:57.269 --> 00:10:01.830 pipeline, or even the sharpest insights. But 00:10:01.830 --> 00:10:05.549 if you can't communicate that clearly, you lose 00:10:05.549 --> 00:10:09.509 opportunities. In interviews, in team meetings, 00:10:09.879 --> 00:10:11.720 So let's say you go into these team meetings 00:10:11.720 --> 00:10:14.519 unprepared. You have the cleanest data pipeline, 00:10:14.759 --> 00:10:17.759 everything. You have solved the solution in your 00:10:17.759 --> 00:10:21.259 head. But if you don't present those correctly, 00:10:21.519 --> 00:10:24.279 what's the use? You're going into a client demo. 00:10:24.860 --> 00:10:29.620 Your ability to explain what you did, how you 00:10:29.620 --> 00:10:33.100 did it, and why it matters is really everything. 00:10:33.279 --> 00:10:36.100 That's how you can close deals if you're not 00:10:36.100 --> 00:10:39.659 involved in that. How do you satisfy your client? 00:10:40.519 --> 00:10:42.360 Because if you're not able to explain what you 00:10:42.360 --> 00:10:46.799 did, they're not going to be happy with you slash 00:10:46.799 --> 00:10:53.019 resign with you, right? And this app is about 00:10:53.019 --> 00:10:57.320 reducing that friction. It gives you a strong 00:10:57.320 --> 00:11:00.419 foundation to build from, like a head start. 00:11:01.480 --> 00:11:05.919 And even more, it brings storytelling into your 00:11:05.919 --> 00:11:10.629 workflow. It encourages you to think about the 00:11:10.629 --> 00:11:13.970 narrative while you're still in the Jupyter Notebook. 00:11:15.429 --> 00:11:19.210 That's great, right? So let's look at some real 00:11:19.210 --> 00:11:21.669 -world use cases of how you can end up using 00:11:21.669 --> 00:11:26.769 this. So let's just say you're a student who's 00:11:26.769 --> 00:11:29.070 working on a Kaggle project. So this is something 00:11:29.070 --> 00:11:30.889 I mentioned at the beginning. Maybe you're just 00:11:30.889 --> 00:11:33.929 starting out in your journey, trying to figure 00:11:33.929 --> 00:11:35.789 out what to do, right? And you're a student. 00:11:36.779 --> 00:11:39.000 finding some cool data sets to work with so Kaggle 00:11:39.000 --> 00:11:41.899 data set you get I mean you use this Kaggle data 00:11:41.899 --> 00:11:45.419 set to analyze what you're trying to do or analyze 00:11:45.419 --> 00:11:51.679 the data set and you produce some results but 00:11:51.679 --> 00:11:54.940 now you are able to turn your notebook into a 00:11:54.940 --> 00:12:00.980 job interview ready presentation and another 00:12:00.980 --> 00:12:05.360 use case could be a data analyst who can prepare 00:12:05.360 --> 00:12:11.240 stakeholder presentations faster without copy 00:12:11.240 --> 00:12:16.379 pasting plots all day right and a machine learning 00:12:16.379 --> 00:12:22.679 engineer can generate internal tech talks explaining 00:12:22.679 --> 00:12:27.279 their model pipeline using real diagrams that 00:12:27.279 --> 00:12:32.500 you get out of this so those are some real examples 00:12:32.500 --> 00:12:36.610 which i mean real world examples which are applicable 00:12:36.610 --> 00:12:39.750 for this right like this is how it can be used 00:12:39.750 --> 00:12:45.830 in the real world um so like i already i already 00:12:45.830 --> 00:12:47.809 gave you like an indication of where i want to 00:12:47.809 --> 00:12:49.950 take this app so i just want to talk about that 00:12:49.950 --> 00:12:53.889 what are the future plans for this so i am not 00:12:53.889 --> 00:12:56.009 stopping here and i hope you are you are not 00:12:56.009 --> 00:12:58.950 as well i hope you're trying to use this with 00:12:58.950 --> 00:13:03.340 me um a link to The blog I've already written, 00:13:03.460 --> 00:13:06.639 which has the full code, will be pasted in the 00:13:06.639 --> 00:13:08.840 show notes. And here's what I'm building next. 00:13:09.419 --> 00:13:12.779 Support for more diagram types like mind maps, 00:13:13.039 --> 00:13:16.360 Venn diagrams, org charts, organization charts, 00:13:16.480 --> 00:13:19.940 right? And connecting with napkin .ai or diagram 00:13:19.940 --> 00:13:25.299 GPT for more rich illustrations. Integrating 00:13:25.299 --> 00:13:28.340 with Google Slides so you don't even need PowerPoint. 00:13:30.720 --> 00:13:35.019 editable slide themes layout presets presets 00:13:35.019 --> 00:13:41.960 also trying to add voice narration generation 00:13:41.960 --> 00:13:47.039 for each slide it's like a voiceover and if you're 00:13:47.039 --> 00:13:49.179 listening and want to contribute let's contribute 00:13:49.179 --> 00:13:52.879 let's collaborate right like this project is 00:13:52.879 --> 00:13:54.919 open source i mean you'll have access to this 00:13:54.919 --> 00:13:59.159 um the code the code will be very easy for you 00:13:59.159 --> 00:14:01.100 and if you were able to build something cool 00:14:01.100 --> 00:14:04.759 with it which i wasn't able to just let me know 00:14:04.759 --> 00:14:08.879 it'd be great for us to collaborate um you know 00:14:08.879 --> 00:14:11.259 this is i'm really excited for where this is 00:14:11.259 --> 00:14:14.139 going and there's a lot of interesting opportunities 00:14:14.139 --> 00:14:21.600 here so to wrap it up this isn't just about automating 00:14:21.600 --> 00:14:27.980 slide creation this is about making it easier 00:14:27.980 --> 00:14:34.000 for all of us to share our value to share the 00:14:34.000 --> 00:14:39.139 value of the work that we're doing so if you're 00:14:39.139 --> 00:14:43.639 a data scientist a data analyst an ml engineer 00:14:43.639 --> 00:14:48.840 or even a student just starting out the ai space 00:14:48.840 --> 00:14:52.539 right if you've ever felt like you know you're 00:14:52.539 --> 00:14:56.559 stuck here explaining your project this tool 00:14:56.559 --> 00:15:00.820 is for you and you can find the code like I mentioned 00:15:00.820 --> 00:15:04.080 in the blog post which is which will be in the 00:15:04.080 --> 00:15:09.500 show notes the streamlit app screenshots you 00:15:09.500 --> 00:15:12.519 will see in that blog post as well and there's 00:15:12.519 --> 00:15:14.399 a full walkthrough in my in that blog post so 00:15:14.399 --> 00:15:18.679 I'd love to hear how you use it or what features 00:15:18.679 --> 00:15:21.320 you want next so just feel free to comment on 00:15:21.320 --> 00:15:27.620 that you can reach me on uh the same blog post 00:15:27.620 --> 00:15:29.500 it'll be like a medium blog post if you don't 00:15:29.500 --> 00:15:32.820 have a medium account you'll also uh get this 00:15:32.820 --> 00:15:36.440 on substack substack is free if you don't need 00:15:36.440 --> 00:15:39.200 a paid membership for that um thanks for tuning 00:15:39.200 --> 00:15:42.539 in and until next time just keep building keep 00:15:42.539 --> 00:15:44.759 sharing and let your work speak