1 00:00:00,000 --> 00:00:05,760 Hey everyone, my name is Mukundan Sankar. I'm the host of the Data, AI, Productivity and Business 2 00:00:05,760 --> 00:00:10,960 with a little personality podcast. We're going to dive into something really exciting. It's another 3 00:00:10,960 --> 00:00:16,560 AI-powered solution and this time I'm applying it to PDFs. Okay, so let's dive into something that 4 00:00:16,560 --> 00:00:22,160 I think is going to resonate with anyone who works with data and AI, PDFs. They're the lifeblood of 5 00:00:22,160 --> 00:00:27,520 our field, but who actually has the time to read all of them? That's where this podcast gets really 6 00:00:27,520 --> 00:00:34,400 interesting. It's an AI-powered tool that I built here that basically turbocharges your PDF game. 7 00:00:34,400 --> 00:00:41,040 It can extract key information, turn it into audio and even generate Q&A. That's not all. It uses some 8 00:00:41,040 --> 00:00:46,880 familiar tools in the data science world like PyPDF2 and Hugging Face. So I'm curious about 9 00:00:46,880 --> 00:00:53,760 how much of a pain point PDFs are in your work. For me, it's a major one. In data science and AI, 10 00:00:53,760 --> 00:00:59,360 staying current means sifting through a mountain of PDFs. And it's not just about keeping up. 11 00:01:00,080 --> 00:01:07,200 Sometimes I really need to dig into a specific paper for a project. Even skimming for key 12 00:01:07,200 --> 00:01:13,120 takeaways can take forever, especially with all the technical jargon. That's where I believe this 13 00:01:13,120 --> 00:01:21,120 tool can be a game changer. Now let me break it down for you how this works. If I start with the 14 00:01:21,120 --> 00:01:29,920 first step, it uses PyPDF2, which I spoke about. PyPDF2, let me talk about that briefly. So PyPDF2 15 00:01:29,920 --> 00:01:36,320 helps you to extract all the key text from the PDF that you uploaded. A lot of people might be 16 00:01:36,320 --> 00:01:42,480 familiar with this PyPDF2, especially if you're in the Python world and the data science world. So 17 00:01:42,480 --> 00:01:49,680 it's a, and those of you who aren't, it's a Python-based library specifically designed for 18 00:01:49,680 --> 00:01:59,200 working with PDFs. But the real magic happens in the next step. Summarization. The tool that I built, 19 00:01:59,200 --> 00:02:04,640 it uses a summarizer from Hugging Face. And for those who aren't familiar with Hugging Face, 20 00:02:04,640 --> 00:02:12,080 it is a platform with a ton of pre-trained AI models. So you can just use any AI model, like 21 00:02:12,080 --> 00:02:19,440 text to video, text to audio, and all that would be available there. And this summarizer that I 22 00:02:19,440 --> 00:02:26,720 built, yeah, you can obviously generate summary too. Summarizer for this particular use case, 23 00:02:26,720 --> 00:02:34,800 it condenses those pages of information into concise digestible chunks. So imagine being able 24 00:02:34,800 --> 00:02:40,400 to grasp the core concepts of a complex research paper without getting lost in the details. 25 00:02:40,400 --> 00:02:46,640 And for those of you who are always on the go, get this. The tool can convert those summaries 26 00:02:46,640 --> 00:02:52,800 into audio files using Google Text to Speech. That's amazing, right? Have you ever used Google 27 00:02:52,800 --> 00:02:58,320 Text to Speech? It's again another Python-based library, but like as the name suggests, it 28 00:02:58,320 --> 00:03:06,400 converts text to audio. And think about how you can use text to audio for your use case, for 29 00:03:06,400 --> 00:03:13,600 anything, right? Like you're on the go, maybe you're driving somewhere and you don't have access to 30 00:03:13,600 --> 00:03:18,240 maybe like your phone or you just like have access to a Bluetooth or something, right? 31 00:03:19,280 --> 00:03:24,640 You are listening to something on the podcast and this podcast is something that you created. 32 00:03:25,760 --> 00:03:30,240 That's where Google Text to Speech comes in. So basically anything that you've written, 33 00:03:30,240 --> 00:03:35,920 you're converting that into audio. So you are listening essentially a podcast made by you 34 00:03:36,480 --> 00:03:44,160 while working out. Another company that does this these days is Google again on their Notebook LM 35 00:03:44,160 --> 00:03:51,680 product. So they are converting your regular text to a podcast. But yeah, this is a bit different, 36 00:03:51,680 --> 00:03:55,760 but this is something that you built yourself. So there's like a different pleasure in that. 37 00:03:55,760 --> 00:04:00,880 So just coming back to this, I think the possibilities are pretty exciting with this tool. 38 00:04:01,760 --> 00:04:08,800 It's like having your PDF on the go, but you are listening to it. Like I said, 39 00:04:10,000 --> 00:04:15,440 it gets even more interesting. Another function which I really want to talk about 40 00:04:15,440 --> 00:04:24,000 is hugging faces, question and answer features. So basically it is used to generate questions 41 00:04:24,000 --> 00:04:32,080 and answers from the PDF content. Now imagine preparing for an important meeting or presentation 42 00:04:32,080 --> 00:04:38,000 and you could have AI highlight the key points and even anticipate the questions 43 00:04:38,000 --> 00:04:44,560 your audience might ask. Now what's fascinating here is that it goes beyond just summarizing 44 00:04:44,560 --> 00:04:49,440 the information. It's creating a more interactive way to engage with the content. 45 00:04:51,680 --> 00:04:58,000 Now I built out a front end application using Streamlit, which is a Python based library, 46 00:04:58,800 --> 00:05:06,000 but it will help you to see the content in a more interactive way. So I'm going to go ahead and 47 00:05:06,000 --> 00:05:12,240 show you how it works, but it will help you to see data on a website. So it's like basically 48 00:05:12,240 --> 00:05:18,320 you're creating a website and Python by itself wouldn't have been able to do it, I think. 49 00:05:18,320 --> 00:05:23,520 And Streamlit really makes that process easy. So anybody who doesn't know how to do front end 50 00:05:23,520 --> 00:05:32,800 programming can use Streamlit library from Python to make an application. So that's what I did in 51 00:05:32,800 --> 00:05:40,960 I created an application which is asking the user to upload a PDF and generate a summary, 52 00:05:42,080 --> 00:05:48,480 generate questions and answers and generate even audio. What is audio you might ask? 53 00:05:49,680 --> 00:05:53,360 So what I spoke about before was the text to speech feature, right? 54 00:05:53,360 --> 00:06:03,280 With the text to speech feature, it is converting the highlights of this text into an audio. 55 00:06:04,000 --> 00:06:07,920 And this can obviously extend to other things like basically you can just convert the whole 56 00:06:07,920 --> 00:06:13,600 text into audio. So I just didn't choose to do that, but I think if you wanted that could 57 00:06:13,600 --> 00:06:20,800 be done as well. So all you have to do is just change a few bits of code and the blog post, 58 00:06:20,800 --> 00:06:26,560 which actually has the step-by-step approach will be linked in the show notes. So do keep an eye on 59 00:06:26,560 --> 00:06:33,600 that. So yeah, just think about this, like this PDF is not just about saving time. It can help you 60 00:06:33,600 --> 00:06:40,640 absorb complex information more effectively, right? Imagine being able to quickly grasp the 61 00:06:41,360 --> 00:06:47,600 key takeaways from a dozen research papers on a new technique. So you're learning something. 62 00:06:47,600 --> 00:06:54,720 The other day I was learning about LLM quantization or something. And see, I mean, 63 00:06:54,720 --> 00:06:58,800 I don't even know that topic yet. I don't know how to pronounce it, but I was trying to learn this, 64 00:07:00,800 --> 00:07:08,400 get a summary of this using my tool. And it did provide like a quick summary. So it was nice, 65 00:07:08,960 --> 00:07:13,040 obviously, I think notebook LLM is greater, but I think this is something that 66 00:07:13,040 --> 00:07:17,360 you can build yourself and that gives me the pleasure of using my tool. 67 00:07:18,560 --> 00:07:26,800 I can obviously use more improvements. And I would say just play around with the tool and 68 00:07:27,360 --> 00:07:31,520 think what else you can do it to make it better. And if you have any suggestions, 69 00:07:32,480 --> 00:07:38,800 feel free to send them my way. So it's important to definitely acknowledge that no technology is 70 00:07:38,800 --> 00:07:46,400 perfect. And we know that AI is just a tool, right? It has its limitations, but it's exciting to see 71 00:07:46,400 --> 00:07:53,040 how AI is evolving so rapidly. And it's helping tackling some of the biggest challenges we are 72 00:07:53,040 --> 00:08:03,040 facing, right? Information overload. So obviously, like I said, this is just another AI application 73 00:08:03,040 --> 00:08:12,080 and I'm hoping to keep generating more AI content, like more how tos, like how to do this using AI 74 00:08:12,080 --> 00:08:21,440 and how to do that. So that's something that I get excited to do. Now, you can also think about 75 00:08:21,440 --> 00:08:26,960 how you can do like an AI powered analysis on websites, right? And I just spoke about PDFs, 76 00:08:26,960 --> 00:08:35,920 so we can extend it to websites, blog posts, and even books. It could completely revolutionize 77 00:08:35,920 --> 00:08:44,000 how we learn and stay informed. So how could this tool fit into the workflow of someone in our field? 78 00:08:44,800 --> 00:08:50,080 First and foremost, like I said, it's like a massive time saver. Now, instead of spending hours 79 00:08:50,080 --> 00:08:55,920 on researching every paper from start to end, I could just use this tool quickly to extract 80 00:08:55,920 --> 00:09:01,280 the key findings, identify the common themes, and even generate potential questions to guide 81 00:09:01,280 --> 00:09:08,400 a deeper dive into specific aspects of the research. And that's a mouthful. It's just, 82 00:09:09,040 --> 00:09:13,440 I want to be able to save time, right? And it's just not about research, like I said, 83 00:09:14,560 --> 00:09:21,200 you're preparing for work, presentations, or a new visualization technique. And this tool could 84 00:09:21,200 --> 00:09:26,240 quickly summarize the key points from several articles and white papers and even generate 85 00:09:26,960 --> 00:09:32,720 potential questions that the audience might ask. Plus, it could be a life saver when I need to 86 00:09:32,720 --> 00:09:38,160 quickly get up to speed on a new technology, right? Like if you want to learn something quickly, 87 00:09:38,800 --> 00:09:47,280 a quick summary, yeah, just made your life very much easy. So I just always go for 88 00:09:47,280 --> 00:09:53,520 optimize whenever I can. So I'm always trying to use technology, especially AI these days, 89 00:09:54,320 --> 00:09:59,600 to make my life easier. And I hope you're also thinking on the similar lines. 90 00:10:01,840 --> 00:10:07,920 So just imagine not just your PDF experience is turbocharged, getting key information, 91 00:10:09,760 --> 00:10:14,000 and you're also getting it in audio, you're asking the questions and answers and getting 92 00:10:14,000 --> 00:10:22,640 answers from it all by just using Python based libraries and AI. So this working through 93 00:10:22,640 --> 00:10:29,520 PDFs is like a huge pain point and tools like this would definitely make life easier. 94 00:10:30,640 --> 00:10:35,120 So staying current is definitely something we need to do in the data science and AI field. 95 00:10:35,120 --> 00:10:43,760 So instead of just like going through so much PDF, just make your life easy and summarize them and 96 00:10:45,200 --> 00:10:49,600 build out an AI power tool that really does everything for you. 97 00:10:53,360 --> 00:10:58,960 So where could this go? I would say no tool is perfect. Like AI has its limitations. 98 00:10:58,960 --> 00:11:05,120 That said, it's exciting to see how AI evolves into some of our, you know, evolves to tackle some of 99 00:11:05,120 --> 00:11:12,880 our biggest challenges. And finally, just would like to say as AI continues to advance, 100 00:11:12,880 --> 00:11:19,840 we'll even see more sophisticated tools emerge that will also that will transform how we interact 101 00:11:19,840 --> 00:11:27,520 with information. It's an exciting time to be in this field. So question number one, 102 00:11:27,520 --> 00:11:34,880 so question for you, what PDF would you run through this tool? How would you use all the time that 103 00:11:34,880 --> 00:11:42,400 you saved? For those of us always under pressure to stay ahead, this tool could be a game changer. 104 00:11:44,000 --> 00:11:50,400 But as with any powerful technology, it's essential to use it properly and responsibly. 105 00:11:50,400 --> 00:11:58,640 And it's while it is really super tempting to see AI as a silver bullet, I do approach it as a 106 00:11:58,640 --> 00:12:04,960 compliment to my own expertise. Ultimately, the balance lies in leveraging AI's power with 107 00:12:06,000 --> 00:12:11,680 while keeping your critical thinking skills sharp. And AI challenges us to become even 108 00:12:11,680 --> 00:12:18,560 better thinkers and learners, helping us engage with information actively, question it, and use 109 00:12:18,560 --> 00:12:24,080 it to strengthen our understanding. So I hope you got some value out of this podcast today and I 110 00:12:24,080 --> 00:12:29,280 hope to catch you in the next one. This is Mukundan Sankar. Thank you and I will see you in the next 111 00:12:29,280 --> 00:12:51,280 episode.