Transcription
AI, Artificial Intelligence. It’s a word that comes up in most discussions. If you’re into content, you’re probably using AI to write better. If into IT, coding is now faster than ever. AI generates visuals, music, commercials, even podcasts. I hope you remember this one. And if you’re into investing, AI isn’t just helping with research or with screening stocks, but is also identifying market trends, building risk models, and even executing trades without any human intervention.
We’ll try to learn some of it in today’s session. This video might be a bit long, but I promise you, this will be worth your time. Let’s begin.
So let’s start with the basics. And unlike what many of us think, artificial intelligence, AI, is actually an entire field in itself with lots of specializations under it. The big circle is, of course, AI. And thankfully, we’re not at a stage where AI becomes sentient, as-in it is self-aware, has emotions. Basically, Skynet-wala situation nahi hai.
Then there’s machine learning, which is basically the brains behind AI. It involves a program that uses large amounts of input data to train a model. In fact, you’ll be surprised to know – initial work on AI started in the 1930s & 40s, but it’s only the last 10, 15 years that the pace has picked up, and that’s largely because of the huge volume of data social media generates on a daily basis.
Next is deep learning – which is about mimicking the human brain. It utilizes something called artificial neural networks. Let me not get into this. Generative AI is the part that creates something new. It does that by learning patterns from the input it receives. You & I commonly refer to it as a prompt. The output can be in many forms – text, speech, an image, audio, etc.
And nestled somewhere around Deep Learning and Gen AI are Large Language Models. These are pre-trained models with a very large set of data to solve common language problems. The popular ones include ChatGPT, Gemini, Claude, Bard, Grok, DeepSeek, etc., and now there’s a growing bunch of them that’ve been trained using smaller, industry-specific datasets, which in our case includes research reports, stock information, earnings call transcripts, macroeconomic indicators, SEBI filings, company presentations, and much, much more.
In fact, I’ll be doing a quick demo of one such platform – Provue.ai. It was launched very recently. It’s free, and it’s packed with some really interesting features. So yeah, we’ll go through this as well.
Ok, earlier in the video, I did use this word — “prompt”. A prompt is simply an input or an instruction given to an AI assistant to guide it towards generating a specific or desired output. Now, a good rule for all of us to remember is that – whenever you are passing instructions to your AI model, try thinking of it as-if you're giving directions to another human being. For example, let’s say you have a cook at your place and you tell her – didi, mere liye khana bana do (please make food for me). So basis this instruction you’ve given – there’s absolutely no guarantee what didi might end up preparing. This can be roti, rice, daal, sabji, pasta, pizza, noodles, salad. Kuch bhi aa sakta hai.
From an AI prompt perspective, what this means is – the clearer your instruction, the closer the result will be to what you are actually looking for. In fact, let’s demo this. Let me pick something that I don’t own, Dixon Technologies. OK, let me pull out the last quarter’s earnings call transcript. Download it. Let’s open Grok. I’ll upload the file. And the prompt I’m punching is – write me a summary. I think this is what most of us do, and the model presents an output in line with the instruction received.
But now, let me try a slightly different prompt. OK. How about this? Rewrite the summary in a friendly, easy-to-understand tone such that even a 10-year-old can understand it. All right, so here’s the output, and it’s definitely very different from what we received earlier. Actually, look at this. It says – “Imagine Dixon Technologies is like a super cool factory that makes phones, TVs, fridges, and other neat stuff. They had a big meeting to talk about how they did in the last few months of 2025.” If I go back to the first version. See, this only says – “Dixon Technologies India Limited Q4 & FY25 Earnings Conference Call Summary.” OK, let’s go down, look at something else. Ah, see this — “They’re super smart with money, they owe almost nothing, have 264 crores in cash, use their stuff super well to make money, they call this ROCE and ROE, it’s at 48% and 32%, it’s like getting an A+ in school!” Nice, isn’t it? Aur pehle kya tha? “Negative 5-day working capital cycle, cash balance, 264 crores, gross debt-to-equity ratio of 0.07, ROCE, ROE, blah-blah-blah.”
I hope the essence of this exercise is clear, i.e., prompts play an important role in shaping the output. If this part is clear to you, let’s dive into it and examine how to write a really good prompt.
Okay, so this part is also called – prompt engineering. I’m sure you’ve noticed ads. Some guy in Connaught Place or Cyber Hub. He’s interviewing people. Prompt Engineers. And remarkably, everyone is earning the same — 42 lakhs. That said, what I’m about to explain right now, might not get you that unknown 42 lakh wala job, but apke webinar-wale 99 rupay to bacha lunga.
So, very simply – there are six core building blocks of a good prompt. The first element is task, and this is where you clearly articulate your end goal. So these are basically your action verbs — generate, draw, write, analyze, explain, etc., and as an example, a prompt for let’s say the hospital industry might look something like: “Write a detailed report on India’s hospital sector from an investing perspective.”
Let’s move on to the second block – context. And this is where one provides specific details or background information. Remember, the more context you can provide, better the output will be. And in our example – I would want to focus on areas like the market size, growth drivers, competitive landscape, regulatory environment, the five-year industry outlook, etc.
Thirdly, we need to add examples to our prompt. Any one example is good. Two is better. And in our case, just to help the AI model understand the structure and depth of the report, I’m attaching a Jan 2024 report – yeah, it’s a bit old – this one’s by Prabhudas Lilladhar on the Indian healthcare sector, and even this can serve as a strong reference for the kind of output we’re aiming to generate.
The fourth thing you want to do is to take care of – the persona. What is persona? Well, it’s the assigning of a specific character or identity for the AI model to adopt so that the output is consistent with the assigned role and its associated knowledge. For instance, for our industry analysis – I can ask the model to adopt the role of a seasoned “equity research analyst” covering the healthcare space.
Number 5 is format, which explicitly states how you want the end result to look like. This can include tables, bullet points, specific section lengths, like how I have done here with a 1,000-word limit, and so on.
And finally, there’s the tone – which tells the AI how to communicate and not just what to communicate. The tone sets the attitude and style of the AI's response, which can range from formal to friendly to sarcastic, persuasive, technical, conversational, etc.
OK, now let’s put this all together in the form of a prompt and notice how these 6 blocks have been incorporated. Firstly, there’s the task – write a detailed, investment-focussed report on India’s hospital sector. Then the context – which is basically what the report should cover. Next is examples – for which I have provided some sources, and there’s also that attachment. The persona is that of an experienced “equity research analyst” in the healthcare sector. The output is needed in report format, up to 1000 words with short paras, bullets, and tables. And finally, there’s the tone, for which I went with professional, objective, and analytical, suitable for an institutional investor audience.
All right, let’s put all this on ChatGPT and see how the output comes out. OK, so this is the healthcare report I was referring to. I’ll download it. Now I’ll upload it to ChatGPT. Let me start writing the prompt. OK, let’s click on the arrow, and in a few seconds, we’ll have the output. Oh, nice. So this includes an executive summary, industry size, growth drivers, competition, regulations, financial performance, 5-year outlook, risk, and also an investment thesis. If you don’t know much about the healthcare industry, then I’m sure this is a great starting point. But do remember – the first response need not be complete or sufficient, especially when it comes to investing. And it’s only when you iterate by either asking follow-up questions or by requesting the model to elaborate will you come closer to your desired output.
Yes, all this might sound a bit overwhelming, so let me propose an easier solution. Let’s go to Provue. Click on “Free flow”. And let me write a prompt like how a lazy person will do it. Give me report on India’s hospital sector, include market trend, growth potential, major player, outlook, etc. So something like this. Now, let’s be honest – this is definitely NOT how you & I have learnt to write good prompts. But kya karen, hum to intrinsically kaamchor hain na?
So, as a remedy to our laziness, Provue has introduced this simple yet powerful functionality it calls as “Smart Prompt”. And to show you what happens. See this. The Smart Prompt has studied it, studied the original prompt, and has even improved upon it by inserting a comparison to global standards. This could have been improved even further by adding “context”. I’ll take it up in a different video, but let’s look at the response as well. Yeah, it’s covered everything we’ve asked for plus a little extra. The sources are available, and some additional questions have also been proposed, which is great because I think the biggest problem for regular people when working with AI, and especially AI for investing, is that we often don’t know what to do next. So Provue solves for it, in its own way.
And, oh. I’ve been saying Provue for a long time. I’m sure some of you are thinking – what is Provue? So, Provue.ai, P-R-O-V-U-E [dot] AI, is a recently-launched investing-focussed AI platform that I’ll be referencing from time to time as we go through the different case studies. The good thing is – Provue is India-focussed, made by Indians, made for Indians. It’s still a work-in-progress, and pleasantly, it’s absolutely free. If you’re interested in accessing it, then just hop onto their waitlist. All it needs is an email, and do test out the many, many things the team there is working on.
OK, before we get to the use cases, let’s quickly go through the limitations of an AI model. The first issue, and a rather serious one for us investors, is AI’s habit of hallucinating. In other words, the model’s output can be factually inaccurate, inconsistent, or even nonsensical, as you can see here on the screen. A second limitation are biases, and because most AI models are trained on human content, it can easily incorporate our biases like gender and race. Also, keep in mind, AI models have a data cut-off, meaning they might not be aware of the latest developments, as I observed with Gemini, who says, and listen to this: “You can generally assume my knowledge extends up to early 2023.” So watch out for recency. Also, look out for performance issues. Platforms like Deepseek are generally slower; I often get those “server busy” errors. And lastly, a few platforms like ChatGPT, they have this thing that as the responses get longer, it tends to optimize for breadth rather than depth. What this means is – while the early paragraphs are often very rich, very detailed, the later sections become a lot more generalized, often summarizing the text rather than offering sharp responses. This isn’t a flaw – the model might be doing some token budgeting. But because we want nice, good, meaty investing answers, instead of ChatGPT, I would prefer Gemini, as the responses there are a lot more comprehensive and robust.
So, these are some of the limitations with AI, and now let’s move on to the use cases. The first use case is, of course, education. You can use AI tools like ChatGPT, Grok, Gemini, etc., as your personal tutor to gain a deeper understanding of financial concepts and investment strategies. Let’s do one on financial concepts, and something that many of us struggle with, and some aren’t even aware of this, is this concept of free cash flow. So here’s my prompt. I’ll suggest you this: pause this video now and examine this from the perspective of the “6 building blocks of a good prompt” – that we discussed earlier. As we can see here, the response is pretty comprehensive. This covers many of the things we included in the prompt – the definition, importance, formula, limitations, etc., and also some additional but useful information like comparing FCF with other metrics and even tips for non-finance readers.
OK, now for a better understanding of investment strategies, let’s examine Provue.ai. The platform has put together a number of investing greats, and while I was spoilt for choices, I went with Peter Lynch. Now, notice how Provue offers a list of question prompts to help you iterate the response to arrive at a better output. OK, let me click on “how do you apply Peter Lynch’s investing wisdom?” Hmm, so that’s 5 points. OK, let me ask a question: How do I identify stocks with high growth potential? Aah. More textual gyan. How about this one – which financial metrics should I focus on first? Ah, that’s better. You know what, let’s convert this into some real numbers. Can you convert this into a screener query with actual numbers? For example, revenue growth of 20% or more, etc. There we go. This is much more practical. So see how the iterative process is working here. And now that this has made me curious, let’s do this final question – Could different industries need different numbers? Okay. So the response does establish that different sectors do need a slightly different screener query as per Peter Lynch’s methodology. My point is – you can do a lot here to further your understanding of financial concepts, and this module by Provue seems like having a private chat with an investing legend. So do try it out. Do iterate on it. And I’m sure you’ll find it very, very useful.
OK, use-case number 2 is – screening, i.e., this filtering of stocks from a large universe. Given how powerful these AI models are, this should come as no surprise to anyone. However, I would still request you exercise caution because some of the information, the response offered by some AI models might be a bit outdated, and in some cases, it might be inaccurate as well. Anyways, let’s get back to it and say, for example, we need a list of companies that are growing their sales by over 20%. Now, this is information that any screener can fetch for us, but what if we are only looking for monopolies with a sales growth of 20% plus? Now that’s something a screener would struggle with, which is exactly where an AI model comes in. So let’s type it out on Grok: Compile a list. Monopolies. Present 20 companies. Table format. OK, so we have an initial commentary – assumption or limitations that the model faced. And here we have the list of companies: IRCTC, Hindustan Aeronautics, Coal India, Pidilite, Asian Paints. So 20 companies with a strong market share, and also notice the column to the right where I’ve asked for some commentary, and although I don’t agree with some of it, it’s still a decently good addition to have within our output.
Next up is market news, market analysis, and AI models can help you stay updated with all of that. A question can be as straight-forward as asking – what are analysts saying about TCS nowadays? And the response here presents us with commentary from various outfits like Kotak, UBS, HSBC, Nuvama, Jefferies, etc. The model also offers a nice summary in the form of sentiments, and as one can see here, it had to go through dozens of tweets and websites to compile this information, saving people like us a lot of time & efforts. By-the-way, this functionality is available on Provue as well. Just look out for the card that says “analyst calls”. And just out of curiosity, I even enquired – “why is Citi recommending a sell on TCS?” for which the model offered me a pretty acceptable explanation. I’m sure I’ve said it earlier, but there’s no harm in repeating it. People often get confused on what question to ask next. Pleasantly, Provue solves this with what they refer to as the “magic tool”, and when one taps on this, the tool proposes a set of relevant questions that might be of interest to you. In this case, follow-up questions such as the impact of ratings, prioritizing of downgrades, and factors leading to “rating changes amongst peers” were proposed. My point is – you don’t need to get confused at any point when using a platform like Provue, and a combination of smart prompts & the magic tool does go a long way in ensuring you arrive at the right response to whatever you are seeking.
Another popular application of AI models is to help us analyze a stock. I’ll demonstrate a couple of ways of doing this. The first is with Provue, and the platform already has a number of companies in its repository with more being added on a weekly basis. Feel free to explore these cards. However, because I was in testing mode, I went with their “Free Flow” option, i.e., I went with a company that’s not there in their directory. My pick was SG Finserve Limited, and let’s quickly go through the interaction I had. So I started by asking which metrics matter when evaluating SG Finserve, and the reply listed about 10 of them – covering its financials, balance sheet, market position, and the overall strategy. I followed up with another enquiry asking the model for a brief on SG Finserve. Again, the response was sharp and useful. Something specific this time, and the model attributed the company’s sharp growth in revenue and profits to factors like diversification, the growth in loan book, a wider client base, and operational efficiency. My point is – a strong, effective analysis of a company will require you to ask multiple questions, and a platform like Provue or any other platform, having all this information within it can be very useful. That being said, let’s also look at an alternative, and I’m going to use NotebookLM this time. So I’m here on my login. Let me tap on the box which I created last month for SG Finserv. Right, so please look at the left sidebar, you’ll see a number of files there. These are the sources that I have uploaded in my notebook. It includes the presentations, quarterly investor presentations, earnings call transcripts, credit rating report, annual report, and also a YouTube video. So with the sources in place, I can now ask multiple questions. So something like – what’s the company’s business model? And here’s the AI model’s response covering different aspects of the business. Notice, how I can also peek into the source. So when I hover on top of 47, I get to see the exact wordings used within the source document. Of course, I can ask more questions. So something like – how does the company manage to keep its GNPA levels at zero? Basically, there is no end to the number of questions I can ask here, and the more you iterate, the better, the more detailed will be your output. So these are a couple of approaches I’m using. There’s no one perfect solution, so do try a bunch of these things, and I’m sure all these will get a lot better over time.
The next use-case is the fundamental analysis of a company. I know we have discussed variations of this already, so I’ll just give you the prompt. Okay – this looks a bit excessive. But let’s break this down and see what all it covers. First up, we have the persona of India’s top equity research analyst. Then there’s the task of conducting a 360-degree fundamental analysis of Solar Industries India – that’s the company I’ve chosen for this. Let’s have a concise overview of the company. The business model. Any economic moat the company enjoys. The competition & specifics of it that Solar Industries faces. The growth drivers, very important. Any guidance offered by the management in concalls, presentations, etc. The key risks and also red flags like share pledges, related party transactions, contingent liabilities, etc. The promoter and quality of management. Again, a very important requirement, this includes any corporate governance issues in the past. Then we need a financial deep-dive covering revenue, profits, margins, the metrics, ROCE, debt-equity, free cash flow, etc. – this would be a meaty section. Then comes the valuation. The outlook for the next 3 to 5 years. And the prompt ends with how we want the output to come out. And yes, there’s a sort-of regulatory fence that says – please do not offer any recommendations. Yeah. So, it’s quite a sizable prompt. I could have also requested for a cashflow model, DCF, etc., but thoda zada hi ho jata. Anyways, it’ll be great if you can try out this prompt on different companies and across different AI platforms to see which one seems more likeable to you. As for the prompt itself, I’ll paste the link in this video’s description, so just extract it from there and try it out.
OK, this video is getting a bit stretched, so let’s quickly run through some of the other use cases. There’s technical analysis, and although it’s an area where my own capabilities are a bit shallow, I do believe most AI models can perform technical analysis and offer a satisfactory output. For instance – this here is a price chart of Solar Industries. And this here is the response I received from ChatGPT based on technical parameters. For people who use charts, I’m guessing this can be very useful. And here’s the thing – instead of you having to go through hundreds of charts, maybe a better application of AI is in helping you set up price alerts. For example: If your preferred RSI is 20, you can tell the model something like – “hey, give me a list of companies which are monopolies and where the RSI is under 20.” Try it out. I think this should work. And if you have a solid prompt for technical analysis, do let me know in the comments section below.
Another interesting application of AI is how it can be used to build or refine your investment strategy based on something that may happen over the next 6 to 12 months. For instance, with everything that’s happening, let’s assume the price of oil is likely to go up over the next 6 to 12 months. This means, you can have the AI platform draw up scenarios leading up to which stocks are likely to do well and the ones that won’t. Another variation is to have AI research on an upcoming trend. Like when I was researching Dr. Reddy’s Laboratories on Provue, I started another thread on GLP-1 requesting the platform to provide information on recent trends & happenings. This led to an understanding of market expansion strategies within India, which led to further questions on what Dr. Reddy’s is doing around GLP-1. Well, from the looks of it, it seems Dr. Reddy’s is truly expanding its portfolio and has even received approval to conduct Phase 3 trials for oral semaglutide tablets. This can go on & on, and believe me, I would have spent a good 15, 20 minutes just on GLP-1. I even researched it on Gemini, and yeah, from the looks of it – GLP-1 can be a big money-spinner for pharma companies, not just in India but all around the world.
The final use case I want to present here, I think we’re at number 8, and this one is on — portfolio analysis. One simple application is for you to extract all your mutual funds, your stocks, PPF, EPF, gold, etc., all the information that’s probably there with you in some excel sheet – just upload it to the AI platform and start interacting with it. What’s my portfolio value? How has it progressed over the years? Which investments have underperformed? Which ones are a bit risky? What changes should I make in my portfolio? Why should I make those changes? – and so on. One can add financial planning as well. I’m 45 years old, I want to retire at 50, my expenses are 2 lakhs a month, let’s take inflation at 7%, etc., etc. I’m not sure how accurate this will be, but I still think it’ll give us a good idea of where one stands, where we ought to be, and what will it take to achieve the desired financial goal.
OK, so this was a pretty longish video. A very tiring one as well. But I truly hope this has given you a better understanding of the possibilities. AI is truly an evolving field, and I have a feeling this is just one of the many videos I might end up producing to this effect. Just to recap what we’ve learnt in this video: Firstly, AI is a vast field, and our current interactions with ChatGPT, Gemini, Grok, and other LLMs are just a small part of it. Mastering prompts is extremely important, and in the process, we discussed the 6 building blocks. One should be aware of AI’s limitations, especially the way it hallucinates. And finally, we explored some powerful applications of AI in terms of education, market news, stock screening, analyzing a company, fundamental analysis, technical analysis, strategy development, and also how it helps you shape your portfolio.
I’d also like to thank Provue.ai for sponsoring and supporting my work and for opening up their platform so that retail investors like you & I can take advantage of this essential technology, and that too absolutely free of cost. So do try Provue. As I said, it’s still a work-in-progress, but feel free to join the waiting list, and I hope you’ll make really good use of it. As always, thank you for your time. I know this was a long video, thank you for your patience. And I’ll see you very soon. Until then.