Section Insights
Introduction to AI-Powered Knowledge Bases
What is the main problem addressed in this episode?
The episode discusses the challenge of accessing and integrating client data from various sources and introduces Adam Sandler's solution of building AI-powered knowledge bases using markdown files.
- Client data is often scattered and unintegrated.
- Adam Sandler offers a solution that simplifies data management.
- Knowledge bases can serve as a gateway to additional services.
- Listeners will learn how to create a foundational knowledge base.
Providing Resources for Building Knowledge Bases
What resources will be available to the audience?
Adam Sandler mentions that a guide will be shared with the audience, which outlines the structure and types of notes for building a knowledge base.
- A downloadable guide will be provided for audience members.
- The guide will help in structuring knowledge bases effectively.
- Seven durable note types will be introduced as a starting point.
- The resource aims to facilitate the creation of knowledge bases for clients.
Understanding Client Needs Through Note Types
What are the key note types for building a knowledge base?
Adam outlines seven note types that help structure client information, including client identity, preferences, project history, and decision-making patterns.
- Seven note types provide a structured approach to client information.
- Understanding client identity and preferences is crucial.
- Documenting project history aids in recognizing patterns.
- These notes help AI systems like Claude to better manage knowledge.
Creating and Organizing Knowledge Base Files
How are knowledge base files created and organized?
Files for the knowledge base are generated based on prompts, and an index file is created for easy navigation and organization of the information.
- Knowledge base files are created automatically from prompts.
- An index file serves as a directory for easy access to information.
- Files can be structured for both machine and human consumption.
- Clients can review the index to verify the accuracy of their knowledge base.
Handling Sensitive Client Information
What precautions should be taken when managing client information?
It's important to handle sensitive client information carefully, including proper labeling and potentially incorporating sensitivity indicators into the knowledge base.
- Awareness of sensitivity in client information is critical.
- Proper labeling of sensitive data helps in managing risks.
- Building sensitivity indicators can enhance data handling.
- Non-technical approaches can still provide significant value in knowledge management.
Transcript
0:00 Every AI build I've ever done hits the same exact wall, and it's always where is the client's data? It's in Google Drive, it's on a local machine, it's in somebody's head, and none of it talks to each other. So, that's the exact problem that my guest today, Adam Sandler from The Bible Edge, is charging almost five grand to solve. He builds AI-powered knowledge bases, otherwise known as second brains, for businesses using nothing but markdown files. No Obsidian, no rag, no fancy vector databases. And the best part is this knowledge base is the foot in the door that lets him sell so many additional projects into the same clients. So, in this episode, Adam is going to walk through the full blueprint live on screen, including the seven different types of notes, the pricing tiers that he charges, and the two skills that keep the whole thing running on autopilot. So, enjoy the episode, and let's dive in.
0:49 Back on the pod, we have Adam Sandler, not the Adam Sandler that you might be thinking of, but what I think is the cooler Adam Sandler, actually. So, he's he's back with us. He's going to share some absolute sauce today on building knowledge bases, AI-powered second brains, how we can make money with that, how we can sell it as a service. So, Adam, why don't you kind of tell the folks in the audience, what are they going to take away by the end of the episode if they listen all the way through?
1:14 >> Yeah, by the end, anyone watching this will be able to build a starter knowledge base with a foundation that will enable it to be grown further over time, and easily repeatable for other clients. and it's such a valuable resource these days because you probably run into a lot of people, or maybe you experience this problem yourself, where you have to repeat things into AI, you lose stuff when a session closes.
1:46 all of the complications around managing context, a knowledge base is going to help you solve. And it's not that technical to get set up, which is why it's such a great thing to use to generate partnerships and relationships. So, really excited about this. >> Awesome. And now, this is something that you're like you're selling this as a service, too, right? I know you're doing a lot of different things >> Yeah. >> as making money with AI, but this is kind of your main like the the bet that you're making right now, right?
2:18 >> So, what I've learned and discovered after working with several clients on different kinds of builds or consulting is that this resource, having a knowledge base, sets you up for so many different ways that you can engage and deliver value with a client. >> Right. >> And what I what I learned was that every time that I'd engage on a build, there was that challenge of like, "Okay, where is the information for this? Where is it kept? How is it maintained?" And that would always be a bottleneck. and so, it sort of occurred to me to start there. When I work with clients or when I first start speaking with clients, and most of the time those prospects have something very specific in mind that they want to achieve, right?
3:02 >> Right. >> Build me a CFO agent, for example. Great, but where is the financial information? Where is all of the context that agent would need to do work, right? And so, answering that question first sets up a project to be much more accessible in terms of reaching those goals. It enables the client to have better access to their information, and it opens the door for future projects because you've already laid down the base for what you can do.
3:36 So, I always am now trying to start with this. >> So, and I love that because we we talk on this podcast a lot about like AI audits and how that's a really good foot in the door. Otherwise, you know, it's sometimes I called it like a tripwire offer. And this knowledge base, I mean, is very similar in the sense that it's kind of the foundation for showing you or for really setting up, kind of teeing up additional engagements with the same client. So, I love it. I think it makes perfect sense. I think it's super valuable. And I know you've got >> of a in the spirit of an audit, you know, and I'm sure you've encountered this as well. When you speak with clients and you get to know know them a little bit, a lot of people that I meet have their information all over the place. Google Drive, local machine. So, built into this opportunity is the prospect of cleaning that up for them.
4:30 And a lot of the clients and prospects that I speak with find that angle very compelling. So, you can say, "I'm going to clean up all of your company knowledge. I'm going to organize it, structure it, and I'm going to make it a living, breathing, tangible asset that can be a resource for anything that you do with AI moving forward. Or >> I love that. >> non-AI partners, agencies, freelancers, becomes a resource that can be shared with any entity that you're working with.
5:00 >> Right, it just makes their business life easier overall. So, I think it's it's really compelling and I I can tell why it's been successful for you. So, I know you wanted to share your screen and kind of jump into some of the frameworks or maybe even a little demo for how you're going about building this and I think people are going to get a lot out of that. >> yeah, there's there's really like not a one-size-fits-all and hang on while I find the share button here.
5:25 and you know, anyone who's spending time reading about AI on X or on social media has likely encountered the idea of the second brain or the Karpathy Obsidian Wiki. These are all versions of the same thing. And I think, you know, there are complex versions, there are simpler versions. And the good news is that you can get a lot of value and capability out of a simple implementation. You don't need to worry about Obsidian, rag, vectors. You can start simple and still deliver a lot of value, which is why it's such a great opportunity for people who are themselves are just, you know, evolving their personal technical capabilities.
6:08 >> I love that. >> Yeah. >> So I've got a fresh Claude code project open in here. And just in case you're wondering, I'm using C mux, which is a terminal open source terminal tool. The reason I like it is cuz it has browser tabs, a lot of other features. and when you are first speaking with a new prospect and want to bring this into the conversation, what I talked about earlier in terms of like, "Hey, we can clean up all of your knowledge. We can put structure behind it." All of that line of thinking of like, "Hey, I'm here to help you get organized and structured." Even with AI out of the conversation, in my experience has been a very compelling conversation for a lot of people because they people are they don't know how to organize it. They don't want to take the time to do it. So that it's a great way to kick off that conversation.
7:00 Another a couple of other great points. We all saw that Fable got pulled, right? >> Yep. >> You saw a little bit of freak out going on online where people are like, "Oh god, if my AI can go away, what does that mean for like my access to essential business resources?" Right. Having a knowledge base that is independent of ecosystem like Anthropic or Open AI helps you with that issue, right? If you're not leaning on Anthropic or Claude to store all of your company information, if Claude goes away tomorrow, you'll be fine, right? Additionally, it allows you to travel to different providers, right? So a knowledge base can plug into Claude as easily as it can plug into Open AI.
7:45 So, you're industry agnostic. we talked about earlier how this is a great foundational starting point. We don't need rag, we don't need all of the fancy indexing stuff, but we can add that in later, right? So, getting in with a client, establishing a simple foundational knowledge base allows you to scale it later in the future and gives you a reason to keep speaking with that prospect, right? And ideally strengthens a longer-term partnership.
8:17 and so, those are a couple of the benefits. Another big benefit that I don't think you hear a lot of people talk about is token costs because if you are providing the same context, if you have one context resource and you're not having to repeat or re-enter things so many times, you're also going to there's going to be cost savings with tokens. And in my experience, especially for teams and enterprise-level clients, that is a very compelling selling point as well.
8:45 >> Well, and that's a good point. And honestly, I'd never even considered that, but it makes a ton of sense because it's like if all the context is stored in one place and you know, theoretically a lot of it might be cached and you it's not having to go and search through 100 different tools to find the answer. It's all right there in the knowledge base. That is going to cut down on token costs. And you know, that if I, you know, my sales brain kind of turns on when I'm pitching that to somebody, it's like, "Look, I might be charging you you know, whatever." And I we'll get into how much you charge and kind of how you structure this offer later, but let's say I'm charging, you know, three grand or five grand to build this out for a client and, you know, maybe a couple hundred bucks a month to maintain it. It's like, "Look, you're going to make that money back in a few months at the latest because of how much you're going to save on tokens because of now how much more efficiently AI is going to be able to retrieve your data."
9:33 Like that is something I never even considered, but it makes a ton of sense. >> Yeah, yeah, especially as I think Anthropic announced their teams I don't know about teams enterprise I think it's going strictly to pay as you go via API. so yeah, so when you are speaking with a client and here's another thing that you'll run into is like like okay, well how do we structure your data? Right? A lot of small business owners or independent operators entrepreneurs are not data analysts. They're not thinking in schema. Right? And they don't necessarily have to be, but you want to think of a general schema or structure for what a client's knowledge base might look like, right? So you want to go in with a starting point of view.
10:20 and what this boils down to are roughly seven different types of notes that can be taken. >> And can you zoom in on those a little more? Is it possible to zoom in? >> Yeah. Let me know if you want me to zoom in a little more there. I know there's a lot of words >> And also too real quick for I'm sure people in the audience are going to be interested like it looks like you you've almost built this into like almost like a wiki style document. Are we going to be giving this away for folks in the audience if they want to download this and and literally hand it to Claude to build out their business plan or whatever they want to do with it?
10:54 >> plan or whatever we're we're we're rushing through the guide for this call, but we'll be sharing this so viewers can go through this in the time and detail that they want. >> Perfect. So then we'll make sure for the folks in the audience we'll put this in the description if you're watching on YouTube and we'll put it in the show notes if you're listening on the podcast. So just wanted to to touch on that. So Adam feel free to continue.
11:16 >> Great. so what I found is that there's seven durable note types or seven note types that seem to always be consistent and I think using these seven as a starting point answers the question for like okay, if I'm working with a client and we have all of this information, what is a starting structure? And I think this is a pretty good one. And so we have a snapshot, who is the client, right? Who and from the client's perspective, it might be who am I? What's my business?
11:47 Who are the key people? Who are the current clients? The current partners that are active right now. Ongoing conversations. Preferences and rules. These are things like, you know, your writing style, how you structure meetings, do you mandate agendas, do you mandate weekly one-to-ones? These are sort of your internal ways of working. Project history. Providing some kind of legacy view of projects that have that are in process or have wrapped up.
12:21 and over time, this is important because AI is very good at pattern recognition. So, when you start working with a knowledge base, a lot of the magic comes in over time. and when you are able to provide more of that history, and there's more opportunity for pattern recognition, you're going to get great outcomes. So, just a little bit of a tangent on that one. Decisions and rationale. So, this one is super useful inside of a knowledge base or when I work with clients and for myself, there's a file that always has the most recent decisions, why we made those decisions, and any other context that might help. And this is super helpful in just a day-to-day.
13:04 If I'm speaking with a client on a phone on the phone on the phone on a on a meeting, and we make a call, you know, I don't have to worry about capturing it in notes. It's in a meeting transcript. It's going to get captured in my knowledge base and filed in a very specific place as a decision. there's open loops. So, these are things that are in flight, anything that is waiting for a decision, for example, critical client conversations, maybe there's a proposal that is in flight and you want to remember to follow up, or there are certain details that you prepare to negotiate. This would be stored at somewhere within the open loop section.
13:41 And then links. And links not necessarily like web links or URLs, but how are we connecting the dots to the source of truth for the information that is being used in the knowledge base. So, hopefully this is clear. Again, the point of this is to you know, it inform some starting structure when you think about organizing a client's knowledge and information. >> And honestly, this is already making it a lot clearer for me because like you said earlier, there's so many different flavors of knowledge bases or second brains or however you want to refer to it. And I mean, just the the way that you laid that out right there, like there's seven types of notes. To me, like that just makes sense. Like, oh yeah, because before that, you know, my head is spinning with like, oh, well, we've got call transcripts and I've got emails and I've got all these different things that could theoretically go into a knowledge base. Like, what do I include? What do I exclude? But this is a really simple framework for determining it's like, okay, well, you know, a transcript we might be able to turn that into a project history.
14:46 Like, if we had a call about a specific project or, you know, a transcript could also translate to an open loop. But, it's just a really, I think a very simple way to kind of quantify what each of those what each of those resources need to be inside a knowledge base. So, I think this is really >> Absolutely. Really clear. >> Yeah, and and one thing I should mention alongside this is that, you know, what I've learned working with clients is it's helpful to identify the spine. And spine is terminology that I learned from working with Claude on knowledge bases.
15:17 and what it means is simply like that one foundational schema. What is that one foundational piece of knowledge or data that everything sort of ladders up to, right? And for a lot of my clients, it is their annual goals or objectives. And the filter for that is like, I can draw a line from every other thing in my business to a goal. >> Mhm. >> Right? And therefore, like, it makes sense that the goal would be like the foundational schema. So, I don't want to get too far into the weeds cuz like it's not necessarily requirement to think about it here, but it's helpful to have that in mind and to be ready to have that conversation.
15:56 >> Got it. Yep. >> And this isn't, you know, gospel, right? I You can sort of put any spin or, you know, take away, add, you know, alter this to your need or or how it makes sense to you. I think a lot of my output is, you know, a product of how I work and how I think and I always try to be cognizant of how clients work and think differently than me, right? So, you want to have some level of flexibility to make sure that you can fit a client's needs appropriately.
16:30 So, when you are working with a client and and you reach out and you find a prospect, it's difficult to figure out where to start, right? But I don't want to jump ahead too much. Are there any questions on your mind before we would get to that point? >> So, no questions, more of just an observation. Like, I like how what you're showing us here on screen, you've got it you've got it almost broken down by module, right? It's like module one is like, here's the offer. Module two is, here's what the, you know, the outline of the actual database itself looks like. Module three is, here's how, you know, step one when you're closing a new client. So, I like how we're kind of taking this step-by-step and this is laid out almost like a little mini course, which again, we're going to give away in the show notes in the description, but it's got a very logical sequence to me and I I want to keep going. This is good.
17:21 >> Yeah, yeah. And And one thing I want to mention like, you know, when we when we talk about knowledge bases or when you hear about them out in the wild, like you might think there's a database, right? You might think that you need to integrate with something. The good news is that you can do this only with markdown files. You don't need to sign up >> Right. On your on your local computer, right? Like you don't need an Obsidian involved.
17:40 >> right? If you're working with a client, you're going to be on their local machine, right? Not >> Mhm. >> You might not be directly working or like you know, remoted into their machine. I direct my clients a lot of the time or I'll create a little like one-time web page where they can just download things from. maybe for another video. but I just want to be clear like part of the reason I think this is streamlined is because like there's no technology. We're just talking about markdown files and organizing them well, right?
18:11 >> Yeah. What what we're not doing is we're not asking a client to go and like install an AI agent or build a Claude scale or, you know, connect to Obsidian or use Claude code or anything like that. It's like you said, we could do all of this that you're showing us with markdown files on their local machine. Like there's nothing nothing fancy about this at all, which is why I think it makes such a compelling thing to sell because the the biggest I think hang-up what I've experienced in selling AI solutions to clients is that is the technical drop-off, right? If the client's not technical or if they think that they're going to have to do something technical, even if they don't, that's where all the friction lies and that's why those deals don't close. This is the opposite of that, which it makes it a really, really good foot in the door offer. So >> Yeah, and I will say it's there have been circumstances where like a client is interested, they just don't have the time, right? And learning opportunities have emerged from that. So some of the clients I work with actually get a lot of value from me just showing them how this works, even if it's simple. So look for those coaching moments as well that could be a real value add for those partnerships.
19:17 So, the first thing that we want to do is like, you know, once we have a client and we're working with them, we want to establish that foundation for collecting and aggregating knowledge. And so, what I've done is create a series of prompts or questions, if you will, and I've done some like created some fictional stuff just to get through these steps. But, we want to start giving Claude, and I'm using Claude code here, but you can do this in Coda, and the direction we're headed here is this is as if we would set up a client in Coda work. with this, right? Because to your point, Corey, like a lot of the less technical people out there are not going to be comfortable with Coda code.
20:00 I just have it up here for the build. Right? And so, we want to once we sort of have like these seven items down, we want to start aggregating information from the client, right? And so, the expectation is that through your initial conversations, you would have done just some discovery and have answers to these questions, right? So, I just created a prompt. I'm just going to paste it in there. it's probably going to be a little janky cuz like it's a fake client and there's no actual resources, but the goal here is that you want to start getting Claude to think about the information for this client and to start identifying where the data connections are. So, having some of this initial input catered for the client, the prompt here is generic. The one I pasted is not.
20:52 This would be you would work with the client to have them enter this on their end in Coda work, perhaps. >> And what are some of the questions here just for the folks that might be listening on the podcast? >> Yeah. So, we're if you remember back to the seven items that we talked about, we're sort of asking questions to get clarity on examples of those items, right? So, the first question here is who are the stakeholders? What does each one care about? Right? So, we're we're adding now more detail to the people and contexts. That would be one of those seven dimensions.
21:28 What does this client care right most right about now? Right? This would be oriented around projects. And notice that one and two are making a distinction between people care and client care because stakeholders and people could be internal people separate from clients. Right? So, through these types of questions, we're starting to establish that detail and Claude is starting to see the distinctions in the patterns, right? We're sort of I don't know if training is the right way to think about it, but it's like conceptually, we're starting to train Claude to see the client through this lens. Not through My owner has been just chatting at me in sessions ad nauseam, right?
22:13 So, there's also the third question is what have we already promised them? What decisions were made and why? What has worked or clearly not worked? So, a lot of this is oriented around project way of working, understanding clients, understanding people. I'm scrolling back up to those seven so we can correlate them. Decisions and rationale, right? Looking for patterns in decision making, understanding what's in flight. and over time, all of these help Claude to recognize patterns and get better with managing your knowledge base.
22:46 >> Right. >> Is that clear? >> Make Yeah, perfectly clear. Make sense. >> So, I'm punching in the the prompt here and we have a response. You want me to make this bigger? Is that going to help you? >> Yeah, if you would. >> Yeah, is that is that good? >> Yeah, a little bigger if possible. Yeah, that's better. >> Okay, cool. So, there's some dummy files in here and just from that prompt, we Claude has looked at stakeholders and found some risk managers and some other elements according to the prompt that we just put in, right? So, now it's starting to understand who the client is.
23:24 And from here, in a real-world scenario, there would be more of this, right? Like you want to try and sort of extract as much as you can through this initial conversation, but this is enough to get started with. >> Well, now Adam, something that I that I comes to mind for me is like theoretically, we could spin up a voice agent to do this kind of initial fact-finding exercise with the customer with the client, right? >> Absolutely. And yeah.
23:49 >> Yeah. >> Go ahead. >> No, no, no. I I I got excited because you reminded me of an idea that I I use a lot now and maybe I mentioned to the mentioned this in one of our earlier conversations. when I'm working with clients or or coaching students, my onboarding process now, I just send them a prompt. >> Mhm. >> And the prompt leverages the Claude ask user question tool. >> Yep. >> So, it just you know, handholds them through a guided experience and at the end gives them a blurb to copy and email back to me.
24:26 >> Yeah, I love that. It's very similar concept. But yeah, it's it's again, it's like how can we have the AI do the you know, the interview portion or at least guide the client to where as much of this is self-service as possible? It's really kind of what I'm getting at here. >> One other method and and maybe this is for another video. I said I used Claude managed agents to create an onboarding agent. >> Nice, yeah.
24:48 >> deploy that on a per client basis and you just replace one context file. >> Right. We need to we need to do a a dive on >> >> on managed agents, for sure. >> So, so the the the the idea here, and we're moving kind of fast, right? The idea here is that we're starting to identify what that information is, and then in the next step, we're going to apply a prompt that is designed to start building this into schema, right? So, like we've identified the information, we started you know, labeling it with what it is, but now we need to structure it.
25:23 Right? So, imagine that this is happening on the client's machine, and you might be copying and pasting prompts over, which sounds very scrappy, but it's a way of working that works. And there's ways to present it that look professional, which again, maybe time for another video. so I'm going to copy this, and there are some placeholders in here. Claude is doing that thing where it consolidates all the text. So, I'm just going to assume that it's going to be able to put in the placeholders on its own.
25:58 So, you can see that in this prompt, it's doing two things: creating a knowledge base with seven markdown files, and those seven files match the seven pieces of information that we talked about in the second section here. >> Right. Yep. >> Right. So, do you see like the method to the madness coming together a little bit? >> Absolutely. Yeah. So, again, to kind of summarize for folks, so it's like we've got seven different context file types that are going to be in our knowledge base. In order to figure out what those, like what's actually going to be in those those files, we do an initial, almost like interview with the client, which again, could be done by a voice agent, we could have it, you know, orchestrated by a Claude managed agent, whatever we want, whatever direction we want to go, or we could just have that conversation ourselves, and then take the transcript.
26:46 You're essentially taking the transcript, or their answers to those questions, and then running it through this prompt right here, and Claude is then going to go in and pull out the answers to or really start to scaffold like those initial context files that are one of those seven types that you gave us there. And that basically the exercise up to this point is kind of what creates the foundation for the knowledge base. Is that correct? >> Exactly. Yep, that's exactly right.
27:13 Claude did not catch on with the placeholder, so I'm going to see if I can just very quickly tell it to use the Halcyon. Use the Halcyon material in this folder. normally, you know, we would be more diligent for for the sake of time we want to get through this. >> Of course, yeah. >> We've serviced our original data and now we are applying structure and this is going to be the very first foundation of our knowledge base. I'm just scrolling up to where we were before.
27:49 >> And I'm sure we'll get to this too, but and I'm sure this is a later module, but it's like how do we So how do we maintain it after the fact? So if we charge you know, let's say we charge a client well, really what do you charge, right? And I'm sure that's a a separate section to not get too far ahead. >> yeah, you know, and and Corey, your audit knowledge I think is really applicable here because I do think that there's an audit opportunity, right? And and to be quite honest, like, you know, I would absolutely love to get your wisdom on pricing as well and I would say like these are more of like a stake in the ground for as far as pricing goes. I'm in New York. It's a high-priced market, you know, I think there's a lot of variables that would influence this one way or another.
28:34 >> >> The audit piece, right? may it might fit in with the way that you're approaching audits now. We talked about how like, you know, maybe you have a prospect and their information is all over the place. They've got stuff in Dropbox, on local machines and Google Drive, right? And an audit can be in the context of a knowledge base how to pull all of that together. >> Mhm. >> And then how to structure it as schema.
29:00 And then you've done half of the work for the client right? >> and a lot of that is very straightforward because if you can just get an idea of where things live and even just a verbal articulation of what it is, right? Like oh all of my brand stuff is in Notion and my client information is in Google Drive, right? That detail is enough for you to get a schema on paper and to deliver an audit.
29:30 >> Right. So and that that makes a ton of sense to me because when we do like for example when we do our AI tools audits, all we're doing is we're looking at like hey how are you guys doing how are your workflows look today? Okay great here's a few AI tools that you can implement to fix you know these specific workflows or these specific bottlenecks. But then it's like well hey if you want us to install them for you or train you on them then that's an entirely separate engagement that's the upsell. So >> Absolutely.
29:57 >> it seems like that's very similar to what you're doing here is almost it's almost like a knowledge audit. It's like hey you pay me and I'm going to come in and figure out where exactly your knowledge lives today and then I'm going to the deliverable is going to be you know these are all the knowledge sources we need to pull from and this is what the schema would look like. >> Absolutely. >> And theoretically the client could if they wanted to take that and go implement it themselves but my guess is that 80 to 90% of them are like well this is great but now I need you to go the the full mile, right? I need you to go and build this for me and that's where you know a $750 audit or $1,000 audit turns into a like what you have here $3,500 core build or $4,700 premium build. And what I like doing too is I like I would even raise the core build and the premium build. I would increase those prices by $750 because what I would tell the client is that, "Hey, the audit is actually we actually roll the cost of the audit into the build if you choose us to implement the build.
30:58 So, the audit's essentially free. It's included in the build." >> That's I love that build. That's that's great. one other thing I wanted to say is like there's so many ways that you can position a knowledge base. I have a client where we are looking at department by department. >> Yeah. >> So, doing a knowledge base for a single department one at a time and then with the intention for there to be connective tissue and visibility.
31:28 >> Right. Yep, that makes sense. >> You know, that it's just one way that you can approach this and I think there's a lot of different creative ways to position it like that. Okay. So, we are in the process of putting together our initial knowledge base consisting of seven markdown files for a fictional company called Halcyon. And we did that just now. With the first prompt that I put in was emulating an initial discovery where we provide details about the client. And then the second prompt, which was running while we were chatting, took all of that information and created the first files for our knowledge base.
32:14 And in the right side of my screen here is the root folder. So, if I open this up, here's where those files live right now. And if we just randomly open one up here, let me check out the snapshot. We have just a very high-level snapshot of who Halcyon Health is. Again, it's a fictional company, so not much information here. So, over time these files would grow either in depth or information such as the decisions.
32:45 And new files would be created where it made sense. New customer files and all of this would be indexed in an index.html. Right? So, an index file, which is automatically created if you're using the prompts here, is exactly what it sounds like. It's a directory of where the what where these are and >> Kind of like the table of contents almost. >> Exactly. Yep. sorry, I'm just trying to get it open so we can look at it here.
33:16 So, this is the index file and this is really not for human consumption. It's really more for machine consumption. But, you know, HTML is the flavor of the moment. it's also easy to read for us here on the call and it could even be a client-facing asset if you want it to be, right? >> Right. I was thinking the same thing. I was like this is something you could definitely put in front of a client so that they can at least quantify like, "Oh, okay. This is exactly what's in our knowledge base. Yeah, this all looks accurate. Okay, great."
33:43 >> Exactly. And you know, we have we're I'm sorry, these files are named, right? We have a date that last time it was touched, right? So, all of the information that's aggregating here is intended for Claude to know what to get and when. which is sort of like a very rudimentary version of like retrieval augmented generation, right? It's not retrieval augmented generation just to be clear, but conceptually for a client who is not trading in thousands and thousands of documents, having Claude navigate an index file is perfectly sufficient to get the reason, you know what I mean?
34:26 So, you don't need to bring in that technical sophistication at the beginning. And you've set yourself up for bringing it in at some point later if you want to. I'm going to close this. Yeah, it looks good. So we have a very rudimentary knowledge base, right? And it's living on a machine. And again, you know, imagine a scenario where you're working with a client and you are basically feeding the prompts for the client to enter. And I've done this again, it sounds a little scrappy, but it works.
35:03 And without there are other more technical ways to do it. But once the client has this information set up, a lot of the clients that I work with are using co-work, right? They don't want to use co-work. I'm sorry, they don't want to use Claude code. It's a little too technical. And you asked a question earlier, like what do we do over time? How do we maintain this? Right? When you hear about the second brains and the knowledge bases out in the wild, a lot of that conversation entails skills or processes or capabilities that are managing the information in some way.
35:39 Right? Generally with knowledge bases, there's different layers of context. There's hot context, context that is relevant to you right now. There's durable context. Context that might not be relevant right now, but stuff that is evergreen and important. And then there's disposable, which is exactly what it sounds like. Context you don't care about it once it's gone. And we want to make sure that our knowledge base always has hot context that is actually hot, right? You want to be able to access this stuff when it's needed.
36:17 And so there needs to be some kind of ongoing curation to enable that. And there also needs to be some kind of process where raw information is processed and then indexed and stored properly. >> Mhm. >> And these things can be driven by skills organically. And I'm using a lot of words to say this. It's actually a lot less complicated than it sounds. and to start there are two I'm sorry. So let me back up for just a moment. Let's imagine we're working with a client and we're going to go and Claude that client is going to be working in Co-work, right?
37:01 In Co-work, we want it to be aware that we have a knowledge base, right? And so what we do is in Claude settings and I'm sure a lot of people are aware of this. If we go down to Co-work there are global instructions. Sorry, it's a scroll challenged here. These are global instructions for Co-work, right? It's basically your Co-work system prompt. >> Mhm. >> And it's very analogous to claude.md for Co-work.
37:31 Now, Co-work projects also have their own instruction files but this is global. And so in the in the guide here, we have a claude.md or global Co-work instructions that can be copied and pasted into a client's Co-work instance that will tell Co-work what the knowledge base is and how to use it according to the structure that we have set up.
38:02 Is that all clear? >> 100%. Yeah, so that way it's like once we build it that's by baking that into the say the global instructions inside Co-work then that's what essentially tells Co-work where it's located, what its purpose is, when to use it, when not to use it. That's a critical piece. I mean, that without those global instructions, it's not going to know what to do with that information. >> Exactly. Exactly. and this is set up as a claw.md. I think we'll probably fix this to deliver it cuz I want it should it should have a placeholder for the actual folder. So, what I'm trying to say is in this claw.md or co-work global instructions, you want to make sure that it knows who the client is and that it knows where the knowledge base is, right?
38:50 so, we'll we'll I'll make sure that there's a version of it that works for that. This one this one assumes that the file is in the actual knowledge base folder. I can see that up on glance. So, so that is one step for co-work and then the next two involve the skills, right? So, we want co-work to be able to be smart enough to capture information that is relevant for the knowledge base without our client explicitly asking for it every time.
39:21 >> Right. That's going to be the key in my opinion. >> Yeah. Mhm. and so, we have an ingest skill that is designed to enable that. And this one is set up to have an inbox folder in it. So, you can drop files in there or whatever you want. You can set up a schedule task or other processes to drive raw information into an inbox folder. And then this ingest skill can then be set up to automatically process it.
39:55 Right? So, the way that I have that set up for me is the inbox is one source of information. Sorry, I'm I'm I'm little side note here. Inbox is one source. For me, like I use Gmail as a source. My calendar is a source. I have another, you know, prospect list or CRM style resource that is a input source. >> Mhm. >> We want to make sure all of our input sources are known and readable and included in the ingest process.
40:29 So whenever you have >> Yeah, you want you want any of your inputs to flow in there automatically, really. >> Right. So there might be some manual work to name that in the skill. So the skill is aware of it, right? >> That's a one-time thing, right? >> One-time thing. Mhm. >> Yeah. >> Yep. and usually like in practice the skill, you know, I'm working with Claude and never and I very rarely go in and edit a skill manually.
41:01 And then the second skill here is curate, which manages the health of the vault and will once a week go through, look at you can see right here stale notes, things that are contradictory, any open items, facts that are non-attributed, gaps that it detects based on client work and communication. So once a week it'll take everything, process that information, synthesize new or revised notes inside the knowledge base, right? And so over time, you know, you can sort of see that flywheel, that knowledge base is in motion, it's ingesting, it's being curated, and your client is broken out of the session, is able to query information much easier, doesn't have to repeat things, right?
41:57 Doesn't have to type in brand voice guidelines every time or paste them in, and is also not trapped inside of one ecosystem. And you are now equipped to bring more workflow ideas to the table. Right? >> But I think I think that not being trapped in like the Anthropic ecosystem or open AI ecosystem or any ecosystem is probably one of the biggest benefits here. It's like that's a huge pitch to the client. Like, "Hey, look, once we build this knowledge base if Anthropic goes rogue and decides to lock down their models or like whatever, if something crazy happens, we could take this to an open-source model. We could be running using a locally on our, you know, on our local computer. Like there we it's totally portable and I think that's a huge benefit that I don't want to I don't want people to overlook that might be watching this.
42:47 >> Yeah, absolutely. Yeah, it's When you really start to think about it, I I get excited with all all the different idea prospects and like opportunities that arise from this. One thing you should be aware of is handling your sensitive information for a prospect or a client. >> Yep. >> obviously in any scenario. and ideally like you're never actually touching it if you're able to work with a client and direct them and give them prompts and and manage it like a puppeteer almost. It's kind of how I think about it. Less of a concern, but if you are managing client information, make sure that you're labeling it properly. There might be a case for building a sensitivity indicators into the knowledge base itself, which can be done through prompting. You know, so I think it's probably a little bit in the weeds to get into right now, but it's super critical to be aware of sensitivity in client information that you may be handling or working with.
43:50 >> Absolutely. >> you're covered. Especially healthcare clients, things like that, you want to be very careful. and yeah, you know, there's another prompt here that is helpful relationship building, but that's pretty much that's pretty much it. You know, pretty straightforward and again, very non-technical. You know, you're working with markdown files and there's just so much value and opportunity. and you can step up into the more sophisticated knowledge base models from here. You're not locked into any path.
44:29 >> And that was kind of one of the things I was going to ask you is as far as like stepping up from here kind of like upselling the client from here. So, let's say you've you pitch them on a a knowledge base build and you fulfill that for them and they're happy with it and they want to do more when it comes to AI. Like, where do you go from here? What's Do you have like a you know, bread and butter second upsell? Like, oh, well, once we build your knowledge base then you almost always need this.
44:52 Or is it just kind of whatever it is they need at that time? >> Yeah, yeah. I I mean, not like one specific thing, but always some version of the same idea, right? a lot of the people that I work with when this is up and going, they feel like they have their arms around their pipeline better and I do too. Like, you unders- like you you can spend more time thinking about an individual prospect. And that individual prospect's information is right at your fingertips.
45:21 And that's led to conversations around workflows or automation related to selling. Right? >> Mhm. >> like, for example, I have a client who uses RB2 I think it's RB2B that gets specific information about visitors to your website. >> Oh, yeah, I've seen that. That's a really cool tool, yeah. >> Yeah. So, like, you know, that information is piped in along with Google Analytics, right? So, we have all of the data around the website. We have all of the prospect information. All of that is in the knowledge base together.
45:54 And that has just made the prospect of starting new workflow conversations so much easier because we we can just tap right into it because we already did the hard work of aggregating it, collecting it, organizing it, and structuring it. >> That's actually so true. So, that that I just thought about this for a second ago, but like that makes a ton of sense is like once you've built the knowledge base, now you have pretty much all of the all of the client's business context in one place, and how easy would it be for you to go back literally to query the knowledge base and say, "Hey, now that all of our knowledge is in one place, now that we know now that you know, you know, Mr. AI, everything about this company's business, why don't you give us the top three opportunities for us to further implement AI?" And like literally let the let Claude give you that answer based on the knowledge base because what it might find, like you said, it's like, "Oh, well, hey, now that the CRMs connected and the emails are pulling in and the call transcripts are pulling in, we actually there's a lot of times we're finding that incoming sales you know, inquiries aren't getting responded to fast enough. Like the average response time is 22 hours.
47:04 So, you know, really what you guys need to build next is some sort of speed to lead system or speed to quote system or whatever, but that's like that was a huge lightbulb moment for me just now is like, "Well, once you build the knowledge base, you could literally have the knowledge base tell you what's next." And the client doesn't need to set you don't even need to sell the client on it. They're like, "Well, yeah, this is what our data's telling us."
47:25 Like, "All right, let's go build it." >> Yeah. >> Yeah, absolutely. and there's also opportunities for a lot of value add. like for example, you know, maybe your client didn't ask for this, but here's a whole competitive roster and competitive insights in your knowledge base, right? >> Yeah. You can build in a lot of stuff that's just kind of like an over deliver type situation you're saying that they're they're like, "Wow, this has been great." Like imagine what we could get if we worked even closer with Adam.
47:56 >> Yeah, and you can find you can harvest insights as well. So, like you know, I put all of my call transcripts in. Right? So, sometimes I'll go back and be like, you know, how did I do? How am I doing on my calls? You know, am I a good sales person? So, and then I work with a client who is measuring transcripts for coaching and training and managing a call center team. >> Mhm. >> And that also is a source of insights for the marketing team.
48:29 >> Huh, yeah. That's so many insights. >> that was never a thing before, you know? So, the point being like think about those opportunities for value add of things that are now accessible because of this, you know? >> Well, that that call center exam I mean, God, there's so many directions you could go. If you had a knowledge base trained on a call center's transcripts, it's like, well, hey, what are the five most common objections that we're getting? Or the five most common complaints that people have when they call in?
48:58 And let's go to marketing and build those into the marketing campaign. Like, one, let's fix it on the fulfillment side, and then two, let's go to the marketing team and bake that into our marketing message cuz clearly that's a pain point customers are having. So, we can speak to that, we're definitely going to drive more leads. >> What are the down to the what are the words they use? Right? Like, down to the words they use, right?
49:18 >> There's so many things you can do with that. That's I'm I'm sold on you know, I already have like That's the thing is I've experimented with the different knowledge base tools. Like, I've used G brain, I've used the built-in LLM Wiki skill with Hermes. I I haven't touched Obsidian yet, but like I've used Claude code to scaffold some markdown files that I have on my local machine, and they're kind of all scattered. But, this is giving me the inspiration. It's like, I just need to pick one source of truth, build it one time, connect all the tools to it, set up the ingest, and just let it run from there, and I think that that's I mean, anybody listening to this, not only of course you can go sell this as a service, and I think you should, but just do it for yourself. Like, experience that that mental unlock for yourself, and and then you can go around turn around and sell it as a service because you'll be able to have the conviction around what it's done for you.
50:09 >> That that has helped a ton, what you just described. cuz that's, you know, I'm I'm able to to share my personal experience on client calls, and sometimes clients are like or prospects are impressed by how quickly I can find things because of it, you know? And then I'll just show them. I'll be like, "Hey, you know, here's how I did it. What was the first thing that Corey and I spoke about last Thursday? Boom, right there, you know?"
50:33 >> Yep. >> I love it. Yeah. Yeah, yeah. So, lots of you know, and I think we're going to continue to see value here over time, especially as, you know, models get more expensive, more players emerge, you know, there's going to be a I think more value in having ownership over your own knowledge base and enabling clients for that, too. >> And I've even said, I mean, I started talking about this a couple months ago. Like, I think, you know, second brain as a service is kind of what I call what I was referring to it as is going to be one of the hottest AI offers. Like, I think there's going to be a ton of startups I mean, there already are a ton of startups trying to build that solution, but, you know, as a service, especially because all the tools are so fragmented now, and it it is still relatively easy to implement, but most people don't want to take the time to learn it, huge opportunity to sell that as a service and then make money from it.
51:27 >> Yeah, and you know, there there like to your point, there are a lot of startups out there, but it doesn't mean that you as maybe an independent or solo practitioner can't also swim in this pond because you working with a client, getting to know how their business works, you know, seeing how the sausage is made, like you you you get domain knowledge working with your client that is just going to make that you know your delivery so much better than something off the shelf. You know, so don't get swayed by startups.
51:57 It's totally worth pursuing it. >> Yeah, a lot of the startups, I mean like you said they're selling like an off-the-shelf like do-it-yourself tool. They're not selling the consulting that that goes with it and really that's what you can charge more for and I think that's where the opportunity lies for folks like me and you and people watching this episode is the consulting that goes with it and helping them structure it specific to their needs. So that this has been awesome. We're coming right up on time here. So Adam, any parting thoughts for the audience here?
52:26 And again guys, just quick reminder so everything that Adam shared here today including this mini course that he kind of screen shared as he went along, we're going to put that in the show notes and in the description. So you guys can go there, download it completely free. Adam, if people want to work with you or if they want to hire you to build their second brain or if they just want to follow you, where do you want to send them?
52:46 >> Sure. you can find me at viableedge.com or viable edge on YouTube and viable edge on all platforms, X, Instagram. Sometimes there's a the but YouTube and.com are probably the best. >> Awesome. Well, we will make sure we put all those in the description. We'll link Adam's YouTube. He'll be a collaborator on this video, too. So guys, go check out his channel and with that we will be back in a couple of days with a new episode. Thanks, Adam.
53:18 >> Thank you, Corey. This was awesome. Thanks so much.
Summary
- The primary challenge in AI projects is locating and organizing client data, often scattered across different platforms.
- Sandler offers a service to build knowledge bases that utilize markdown files, avoiding complex tools like Obsidian or vector databases.
- The knowledge base acts as a "foot in the door" for further engagements and projects with clients.
- Sandler outlines seven types of notes essential for a knowledge base, including snapshots of the client, project history, decisions, and ongoing tasks.
- The process involves initial client interviews to gather data, which is then structured into the knowledge base.
- Maintaining the knowledge base requires ongoing curation and skills to ensure it remains relevant and useful.
- The knowledge base provides cost savings on token usage for AI interactions and is adaptable to various AI platforms.
- Sandler emphasizes the importance of understanding client needs and tailoring the knowledge base to enhance business insights and operational efficiency.
Questions Answered
What is the main problem addressed in this episode?
The episode discusses the challenge of accessing and integrating client data from various sources and introduces Adam Sandler's solution of building AI-powered knowledge bases using markdown files.
What resources will be available to the audience?
Adam Sandler mentions that a guide will be shared with the audience, which outlines the structure and types of notes for building a knowledge base.
What are the key note types for building a knowledge base?
Adam outlines seven note types that help structure client information, including client identity, preferences, project history, and decision-making patterns.
How are knowledge base files created and organized?
Files for the knowledge base are generated based on prompts, and an index file is created for easy navigation and organization of the information.
What precautions should be taken when managing client information?
It's important to handle sensitive client information carefully, including proper labeling and potentially incorporating sensitivity indicators into the knowledge base.