Unlocking AI: Chatting with Your Documents for Business Efficiency

**[00:00:00]** Hello everyone, my name is Brando Prescar and welcome to the web talk show. Today we are talking about chatting with your documents. This is going to be interesting because a lot of companies, small businesses in particular, want to start using AI, but they don't necessarily know where to start. That's a common thing that I've seen with many different businesses. The the main thing here to consider is you want to do something that is simple for you to start with but that actually helps you out. So one quick thing you can do is set up a flow where you can chat with your document. So we're going to talk about a few things that are related to this, what this means and some easy approaches and some a little more complex approaches as well so that you know what's a possibility within this. So, if you've never seen any of my videos, welcome. And if you haven't subscribed or like, please like and subscribe the video as well so that you can get access to the next videos when I'm going live. Today, we're testing a different setup like always. Uh, now streaming directly to LinkedIn at a higher resolution directly from OBS. So, we'll see how that goes. Now, let me know in the comments if you have any questions. Please put them in the comments so I can see them here and I'll answer any questions you might have. If you have a small business and you're just thinking about this sort of stuff and want to figure out what you can do directly in your business, please put it in the comments and we'll chat about it. So talking with your documents, this is an interesting approach because many times as business owners, **[00:02:00]** we will have some vast library of documents, best practices, uh frequently asked questions. We'll have a knowledge base. um PDFs that we use to reference and a lot of information that has been acquired throughout the age of your business. And many times you as a business owner and many of your staff that has been with you for a long time, they know all this pretty much by heart. But when someone new comes in, these people don't know all these best practices, standard operating procedures, frequently asked questions, etc. So getting them on boarded might take a little more time. So when a customer comes in and asks them some questions, they might not know exactly what to do or what to answer. So that's where this topic comes in, which is chatting with your documents. Chatting with your documents means you can have a chat interface like if you're chatting with someone via WhatsApp or Messenger or or Telegram or whatever, but it could be in your browser. It could be anywhere you want. and you could be chatting with something which basically can give you responses back from your documents. So, as you've probably seen around the internet, maybe on YouTube shorts or Instagram or Tik Tok, etc., Some people talk about chatting with documents and there's very easy ways to do this like just going into chatgpt or claude or perplexity or gemini or any other LLM large language model and just uploading a file and then you can ask it things about that file. That is basically the simplest approach. If you open Chad GPT this is what more pe most people use. So I'll just talk about that one right now. Let's say you just open TajT. You'll see there's a little **[00:04:00]** plus sign and depending on the interface that you're using or you're on desktop or on mobile, you can just click that or tap that if you're in the phone and then just select files that you have. So, if you have a PDF with all your frequently asked questions, you can upload that. If you have um a document with all the services that you offer, you can upload that. If you have contracts that that you need to talk to or try to understand, you can upload that. And so you now have these documents within the context of the chat session. And so you can now ask it questions about it or talk about what each of these things mean. So there are a few different things you can do with this um onboarding that we were talking about. Then also understanding. So I'll start with the understanding step because I know a lot of people have this particular use case. Let's say someone comes and you brings you a contract and you see that contract and it sort of makes sense but you're a little bit confused about some of the terms. Well, if you upload the whole contract into that chat interface, you can start to ask chat GPT or cloud or whoever whatever tool you're using what each section means. So you can go ahead and say what does this table B mean or what are they saying here or what is hidden behind these sentences because as you know sometimes these contracts might be a little too legal or complex intentionally to make it confusing. Some are not but many are and you might want to understand a little better. So, this will not be legal advice, and what I'm telling you is **[00:06:00]** not legal advice either, but it will help you understand it a little better because it can sometimes extrapolate some of the context within the whole document. And that way, even if you've read it through and you sort of understand it, but it's still a little complex, it will allow you to ask questions that are very specific. And then you can also ask it if it's common practice or if it's a little exaggerated or if it's biased, etc. And I've done this before where maybe you get a type of NDA or something like that and you're just wondering, maybe it's not part of your industry best practices or or maybe you're coming into a new industry. You might get something and you're just you just want to be careful and want to make sure before talking with an attorney. Maybe you're going to talk with an attorney anyway and that would be wise, but you you're just you just want to know is is this best practices common practice? So you can just ask it, hey, are these terms correct or do they make sense? And it might give you some surprise and maybe it'll say yes if you look at and then it could reference some other documents on the internet and it could say, yeah, these follow sort of the same patterns. It looks okay. Or you know what, I've never seen anything like that. It looks completely biased to the other side. maybe you want to talk with an attorney and be a little more careful with it. That's just one example of what you can do. And and if you have questions again, please drop them in the comments. So, the the power of these tools and being able to talk to **[00:08:00]** documents is magnificent because you can very quickly get to the bottom of something without having to spend hours reading it. And I'm not saying don't read documents. especially conscious you have to read them but for many things there's a faster way to get to what you need so I'll give you another example even in code so let's say you're a software engineer software developer and someone brings you a piece of code that you have never seen or you are inspecting a code that you wrote five years ago and maybe it's not as documented as you would like and there's an error somewhere well you could try to inspect the whole code and try to understand the whole thing yourself and you will probably be able to if you know programming, but it'll take you a while. But if you instead upload the files, the related files to an LLM like I said right now or even within this is more technical, but if you use VS Code or anything like that, you can use LMS there as well. You can ask it to review the code and explain it to you and say what are these sections? How are they related? And then it'll it'll tell you it'll say well this application is supposed to do this this this and this and then this function calls this other function and uses these variables and this is how it works. And so once it understands it you can say okay great now I have a question we're seeing this particular error that's coming up and we don't know where to start. And so the thing will go through it and say oh it looks like this is what's happening. Maybe there's a null variable where maybe **[00:10:00]** there's something missing. Maybe the variable changed within time and it's an edge case because it never happens but it just happens with older customers that had an account that was created before January 1st or something like that. So it can point you to the right direction even if you don't want the LLM touching your code which is totally fine. You're good either way. It can help you point out where you should start looking and then you could do that pass and figure it out for yourself now that you know where to look. So you save a lot of time. Again, this is just literally just putting information in it so that you can sort of chat with it and get information back from the documents. Now we talked about onboarding as well. So for onboarding, this is another scenario that's really nice where you might have again new employees. So maybe you have a chat session on chatbt for your business where you previously uploaded all the documents like best practices, frequently asked questions, what services you offer, details about all the services, what things you include, what things you don't include, your warranties, um extended warranties, manuals for your tools, your products, your services. If you have all that linked to a conversation, then someone that is new can when they're interacting with customers very quickly talk to that document base and get information about things but tied to your organization's information, not just a random thing that it's trying to get off the LLM's knowledge or memory. it's actually basing it on documents that you already have in your organization. So, it's very powerful. Now, having said that, there are misconceptions about how this all works. Typically, when you use an LLM and **[00:12:00]** you upload a document to it, it will do one of two things. The first is it will just grab the whole document and put it in its context. So depending on the context window, which is let's just call it the the short-term memory of the thing. Depending on its size, all different LLMs have different sizes. If that document is too large, it might go over that context and so it won't have everything there available to really get information from it. For a regular document, it's just fine. Especially with newer Frontier models, their context windows are huge, so you're going to be fine. But there are some scenarios where you're putting a lot of files and it might go over the context. So the second scenario is it will put it into a what's called a vector store. And a vector store is basically a way to map out all the document or documents in a big database that's semantically relevant. So things that make sense in the context of language. So if you're talking about I don't know a hammer, right? Maybe you have you sell drills and hammers and nails. If you have a section that talks about the hammers in your documents, an a document and maybe you have thousands of documents. Well, a vector store will allow you to get to the information about that hammer faster because of how the word hammer or things related to hammer are related mathematically in in numbers. And so don't think too much about the technical aspects, but here's what you need to know. A vector store is really good if you have a very large knowledge base that would be very hard to do a regular search for perhaps it's very quick at getting **[00:14:00]** relevant information related to a query or a question. So if you ask which type of hammer should I use for this particular type of nail or this particular type of job or this particular type of thing that I'm trying to put against the wall and it'll it will very easy very quickly find what's relevant and bring it back and it'll bring back 5 10 20 results of information just little bits of chunks of information that then the LLM will read and generate a response for for you. So, it's called retrieval augmented generation. It means it's not just going to grab the thing and spit it back at you. It's going to generate a response based on those chunks that it found. It will try to use the chunks that are more relevant. And we can get it deeper into that and then give you that information back. So, it's really neat because you can very powerfully get information that's extremely relevant to what you're asking for to a natural language query to a question that's spoken in just just regular language, not not like a database type query. So, that's why it's really powerful. However, there are caveats. If you are looking for part numbers or serial numbers or something that is obscure and not generally documented, then a vector store is not very good because if you're looking for a serial number ABC123, if it tries to look towards throughout the knowledge base, it will try to find things that are relevant in language terms to that. But there's nothing relevant to that in the real world. ABC123 doesn't really mean anything. It just it's just a specific serial number. So it will have a hard time finding that in the vector store. So **[00:16:00]** vector stores while they are fantastic for reading contracts, documents, etc., they are not very good at finding very specific data that's arbitrary like a serial number or a part number. And there's a different approach that I'll talk about in a second. So, it's good to talk about that because not many people talk about the downsides of vector stores like they're just fantastic. Well, yeah, they are very very good, but there are these edge cases. If you know you're going to have to find part numbers or things like that, then you have to have a different type of search. We what we like to do is called a hybrid search. And the hybrid search is a two-tier approach where you do the vector search like we were talking about, but then you also do a regular free text search into a regular database that's like Postgress. And this is very technical, but just so you get a glimpse of it, it could run both things at the same time. And if it runs both brings the results, if it finds with more percentage of accuracy or or a better probability score that it's it's a good result on the left side on the text search, then we choose that one. If it got a better result from the right side or from the vector store, then we choose that one. And then you can go further and add reranking, which is another thing. It's another concept where it'll grab all those results and then it will use a service like go here to grab the results and then rank them again based on the original query. So again very technical but this allows you to get better results from both your vector store search and your **[00:18:00]** regular search for whatever query you're doing and it allows you to get much better downto-earth grounded responses based on your knowledge base. So if you have a legal type knowledge base with legal documents and contracts or financial data or even product data that's based on SKUs and things like that then it's very good to have this sort of approach instead of a basic like we were talking about the beginning chat GPT window just uploading files it won't work as good right so you want to have this more robust approach when you're doing something where either it really matters um or there like there can't be any room for mistake. That's what you want to do. And you can have a lot of different parameters there to make sure that it's giving you the right data. Now, for most people or for for the basic use cases, which we're talking about, the the simplest thing you can do is just open the window like just you don't have to have any database or anything else. Just open a chat window and then upload the documents and you can immediately start interacting and seeing how it actually gives you results from your data. It's basically like having a chatbot that's linked to your data. And that's really nice because you start to see or imagine benefits for your business. Obviously, you can tie it to a chatbot. And this is what a lot of people do. Many years ago, what we would do is we would create chat bots for e-commerce or for even service businesses like like uh dance schools or things like that. And we would have to program the whole chatbot step by step. If they click this, then do this. If they click **[00:20:00]** this and do this, and it would just branch out. it would be this massive tree of logic to be able to give people the the hours for kids of that age um in jazz or something like that and it was a lot of work. Nowadays, you can build that same exact experience in minutes because you can already have all the details just written down in natural language in a document for the classes for who they are, what the ages are, what the times are, who the teachers are, what the requirements are. have all that just put it in a chatbot that understands and then you can just start talking to it and so you can plug it into your website you can have it on WhatsApp on Telegram etc. And again in minutes you have this this sort of agent or chatbot that you can interact with or your customers can interact with and it can actually provide all these answers directly without having to do all that branched out logic. It's really neat. It brings down a lot of what was very complex for a business to do to the everyday person. So even if you had don't have a huge budget, you can start doing something like this. And a lot of tools, even chat tools like intercom now have their own little agent feature which allows you to put in documents and things like that. Voice flow does this as well. And then you just as a business owner can sign up, set it up, put the things there. Domain chat does it as well. and then just just go with the flow and start putting up there and have your customers start talking to it and get all those frequently asked **[00:22:00]** questions answered fast. So, it's great. That's why I like this because it really democratizes the access to very advanced tools that was really not a possibility before. So, it's really neat. I suggest you try it out if you you've been wondering where where can I just put some AI or sprinkle some of it and see what it really could do for my business. That's what I would recommend or one of the things I would recommend just try to start a chat session with one of them. Put some documents there and start start interacting and seeing getting some ideas of what it could do for you. And again also you could just create a chatbot or even a voice agent and have it where people can call in and they it could actually book your appointment. It could give people the responses about your frequently asked questions etc. Transfer to a regular person. We've talked about that a lot in other sessions and of course if you have any questions about that, please shoot me a message. But there's so much you could do as a business owner that with AI that will help you accelerate your business's growth and also provide a better experience for your customers and for your employees as well. Now the decision here then becomes where where do I start? Do I start myself? Do I need a complex infrastructure? Do I need to do this whole thing? My suggestion for you is look at where you're having troubles. What's the bottleneck right now in your business and see if you can optimize that. So let's say your team wastes one hour a day per employee capturing data and moving it from one system to the other, spreadsheets, etc. If that's **[00:24:00]** something that's happening, then that's one hour per employee that you could recover per day. So if if you have five employees, that's five hours a day, right? So it's a lot of time that can be better utilized. So if you see that that is slowing you down, maybe it's stop slowing your delivery for customers or maybe it's just making the team inefficient uh or it doesn't allow you scale. Focus on that. That's the main issue. Then just there's a way to automate that. And please reach out if you have this situation. We can set it up so that these tools and it doesn't have to even use AI for this that I'm talking about. It can grab information from one spreadsheet that somebody captures in Treasury or whatever and then move that over to the CRM spreadsheet or system that somebody else is touching. And then when a payment comes in through the billing spreadsheet or system, it can read that and then cross [laughter] check all of them and grab the information and put it in the summary spreadsheet that the CEO sees and it just avoid having to recapture a lot of the data because when there's the recapture, there's a lot of human error potential. So these workflows even without AI can really help you save time but also reduce human error which is great and it will bring you back time for your employees which allows you to have them do something else that might be more valuable than just capturing data. So I would focus on that just the very first thing that comes to mind where you you're leaking time you're leaking uh resources uh you feel it's not being productive for your business. Another one that I see **[00:26:00]** a lot is speed to lead. So a lot of companies don't have any issues with leads. They get a lot of leads, but they get so many leads they cannot follow up and so they end up going to the competition. So you can set up systems that allow you to contact leads immediately via text, via message, via text and message, via WhatsApp, via Telegram, via voice. You could call them directly and book an appointment. It can send them a a a packet of information. So, if your intake form flow is something like people fill out a form in the website with a few questions and then someone on your team goes look at those and then generates a document for them manually and then attaches that to an email and then sends it out. Of course, it's a bottleneck because they can only do one at a time. But if you have a system that when a form is filled out, it automatically does all that. Generates the document personalized for them and then sends it out, it's [laughter] it's so much better. And nowadays, you can even include video with it, audio, like dynamically generated audio with it. You can call them automatically with the information. You can book appointments. It's amazing. You really can do a lot of things with AI and just workflow automation in general. So, if you are on the fence for what you can do as a small business owner, just think about that. Think about where you are having pain points with time that's not being utilized properly, with things that are stopping you from growth, with things that are stopping you from reaching out to people fast, things that are slowing your delivery for clients, um, finding **[00:28:00]** information. Maybe your business is heavy on finding information and documenting or referencing or summarizing. All that is also something that you can automate. And then there's also emails. When emails come in, maybe you get a lot of spam. You want to sort that out and focus on the things that are important. Maybe you get in between all your emails, you get some invoice emails that you need to process, but they get lost with all the rest of the stuff. You can also have flows that read the emails and when they have a specific subject or when they're coming from a specific client, it should go into the accounting system and go to billing and then go to payment and follow your whole AR flow. There's again there's so much you can do. And if and if and if you're stuck and you're thinking, oh, I want all this, but I just I don't I don't know where to start, then send us a message. You can send me a message here on LinkedIn. You can send me a message via Twitter or X, Tik Tok, Facebook even. And of course, you can send a message on the website and I'll be happy to help you out. There's there's so much we can do. And that's why we do these videos. We we want people to understand what's possible and what's at your reach, what you can do today to make your business grow faster or better or healthier and make it better for the whole ecosystem, for your clients, for your staff, for yourself, for your family. There's so much that can be done. I'm really excited about what 2026 is going to bring to a lot of businesses because it's just the tip of **[00:30:00]** the iceberg. people have no idea what's coming and what's possible and I think it's really exciting, what's coming up next with everything that's going on. So, thank you everyone for joining the live stream and if you're listening on Spotify, make sure to subscribe to the podcast and if you're on YouTube as well, make sure to ring the bell if you like this content and subscribe as well so that you are notified whenever I release more content. I do these live streams often and then also I have some shorts that go out. So you'll be seeing some of those shorts come out in YouTube, Instagram, etc. Tik Tok and all those places as well. If you ever see something that's interested in one of those shorts and you feel like, oh, it's interesting, but it doesn't didn't really cover much. It just sort of told me something quick. Check out the podcast. So if if you're looking at a clip, some they're clips from the longer video. So check out the web show the web talk show podcast on Spotify or on YouTube or iTunes etc. Um and that that's where you have you'll find the longer form videos where I explain things in more detail. Okay. So thanks everyone and I'll see you in the