Building a Personal "Open Brain": Implementation and Future Potential

01We're trying to make this relatable and something that you can go build. And you could go build this today with just a little bit of, you know, stick to itiveness, a little grit, and the ability to follow some of what Nate Jones outlined and then just keeping up with the community. So anybody can build this, you don't have to be a technologist, but this conversation is gonna get pretty geeky. Because there are some there are and so we'll do our best as we go along here, I think Veer, to stop and give people a bit of an idea of what we're talking about in layman's terms. And we're gonna have to watch ourselves a bit because this is this is really cool stuff. You you need to know about it. I think it can accelerate executives' productivity by an order of magnitude or maybe more. Yeah. Two or three orders of magnitude if they get it right. ⁓ Opening line for every one of them is I didn't want to Leroy Jenkins my way into building something.

02So a lot of what I did before going down the path, because building, and we've I've told other people this before, i i you know, is easy in i i it ⁓ in a recipe type of mode. You take the ingredients, you put them in the bowl, you mix them up, you follow the you follow the recipe, and and you can get from point A to point B pretty quickly, right? The question is is did you, you know, did you do the the the cooking stuff like Missan Place, right? Everything in its place, right? You get everything laid out. Did you do that? Did you get all the ingredients? Did you have a good plan? Did you preheat the oven? You know, all those things. And so ⁓ I tr I tried to take a measured approach going into this to really, you know, kind of think it through ⁓ what was going to be important to me. ⁓ you know, you and I talked a lot about the value in our chat history. and extracting the chat history from Chat GPT and from Claude and from Gemini and Mannis.

03And one of my kind of things that I I built into the the the tool and the product that I'm putting together is most of those chats don't deserve to be memorialized. And so I took my chat history and have Run it through this process. I don't want to interrupt you too much, but I d I also don't want to get too far ahead because I think this is what people will will want to hear about. So maybe just walk us through high level, like what what is your what is the architecture? Like what are you what are you trying to do for this database? And then we can talk about some of the nuance, you know, as you see it around, you know, kind of where you were headed. Like this is how I rank order things. Well, ⁓ so we w we do we wanna get into the the the nerdy pieces that both of us are using to build out this or Yeah, but I think I think it would be good. Let's just I mean we can we can kinda go back and forth here, I think, and and s like maybe just start and just like maybe explain your tech stack.

04Yeah, so like right, so ⁓ d the the the core the database where everything's getting stored. For me, I'm using superbase. Superbase, same because that's what now I was already using Superbase. I think it's a little bit different for you because you were using some other tools. But Nate suggested I mean, there is a guide out there for this. It's just as nerdy, if not nerdier, than what you're hearing right now, but there's you know, thousands of people are trying to implement, and that is the suggested tool for the database. And why? You're better to answer why. There is. I well, I mean it's i i it's i it's extensible. And I think because superbase has another component that is directly integrated with it, which are superbase they they're called edge functions.

05And ⁓ that allows us to interact with that database ⁓ and build what we've talked about already, MCP interfaces in a in a way that is ⁓ pretty cost effective, ⁓ because they ⁓ They are designed ⁓ to spin up and be used ⁓ when they're needed, but they they don't run all day long and so you're not paying the ⁓ this ⁓ twelve months you're buying a computer for the forty eight milliseconds that you're using Yeah. I mean the cloud tax is real and maybe we'll talk a lot about, you know, at at a future time about ⁓ you know, the the conversation about repatriation of of things from the cloud back down to the local data center, which is is is is a thing that is happening right now ⁓ across industries. But the idea that you can run things as needed, pay by the SIP instead of, you know, having a faucet running all day long. Yeah. So ⁓ so ⁓ so superbase and their MCP pieces and then ⁓ are are where a big chunk of my stack lives.

06⁓ the other thing that in order for this to work, you need to be able to we talked about build this semantic layer which involves creating the we talked about coordinates, ⁓ you know, the mathematical ⁓ coordinates which is called embedding. And so you need a tool to do that. And ⁓ and I'm using a embedding trained model ⁓ from ⁓ OpenAI. ⁓ and and so ⁓ now this is where I think you did something similar, right? Are you using OpenAI? I used I used so maybe I did, but this is this is great. This is like a good point for you know the people that have been in technology for a long time and then the people like me who haven't and are you know learning as we go because we're curious. I used the PG vector extension. So built into superbase, there's a vector layer. ⁓ okay. So and I'm Well, and I'll I'll interrupt you just so that I'll give like the nerds ⁓ the piece, right? So PG, right? Postgres. Postgres is the database or one of the main database types that that Supa base hosts.

07PG vector is a set is a layer that allows you to do vector embeddings on top of a Postgres database. Yes. So now I understand because I am not as good at this stuff as you. So yes, I am using OpenAI through OpenRouter to do the embeddings and then write back to that layer. Did I say that right? Is well close. ⁓ yes, I dude, I'm telling you that like little moments like that make me feel really good about myself. And I think the fact is is l for for a lot of this technology you don't have to know, right? I mean it's ⁓ y y you know that's not magic and at the same time you know that like, you know, they're they're ⁓ follow the instructions, like Yeah, you do follow the instructions, but at the same time, if you're like me, which is why we're stopping so much here and talking about, you know, defining this, is I want to know what I'm doing. I don't want it to be a black box. So I want to understand the the different components. I also think it makes me a better prompter.

08I don't know that I need to be in six months from now, but it gives me a deeper. you know, understanding of everything that I'm doing and it allows me to have conversations like this and not be completely a fish out of water, even though I don't have twenty five years of IT experience. No. And these layers, I mean, again, there's lots of tools. ⁓ the PG vector pieces in there. Like these are ⁓ they're just the the the building blocks that make, you know, at the end we just want a tool that works. This is just this is this is, you know, part of the part of the deal. And that's kind of baked into the the edge function piece of of Superbase. Yeah. That would be I guess that's another piece of advice I have is stick with this stuff. Stay curious and don't feel like you have to know everything. I mean, when you learn something, you never learn all of human history if you're in a history class. You learn one piece of it. And all those little pieces build on each other. So using this technology every day and having conversations like we're having now.

09That's all gonna just build your knowledge base. ⁓ it can feel intimidating at first, but I wouldn't go down a rabbit hole trying to understand like how, you know, PG Vector started and who created it and you know, all of that stuff. But I mean, like that is a that's a temptation as you as you build these things. One of the things as we talk about building these products is the fact that there is a struggle here to not over optimize or over-index on on on like knowing how it works, because some of the value in the what we're talking about is building something that you actually use. And if you use it, it will start to give you value. I mean, ⁓ it's a very ⁓ I I wanna say it's a natural law kind of a thing. This is like learning to run a marathon, you don't do it in two weeks. You you you build up to it and and you have to build the the the the endurance and and so it takes time. It does. It takes a lot of time. Okay. So we've only got through one part of the stack. So we're using superbase. PG vector, edge functions.

10So a lot of the capabilities of that cloud SQL database. That's where everything ends up, right? And then there's two other kinds of things, ⁓ or really it's ⁓ there's the s ⁓ a a way to look at it ⁓ is called a capture surface. There's a surface through which the pane of glass through which you are allowed to interact with this information. ⁓ for me, my inputs are mainly going in through Telegram, right? Which is ⁓ you know, ⁓ just a chat tool like like WhatsApp. Or ⁓ Slack or Teams. Yeah, I used Slack but I also already created a manual endpoint that I can then use. I I'm thinking about having it write a very tiny like app that I can just execute on my computer and then have it sitting at the bottom of my computer all day long. Thank you, 'cause that's like d duh.

11⁓ I I I I I I I I I was sitting the other just just the l last night, I I was just thinking to myself, ⁓ I wonder if there's like a ⁓ like a like a widget or something that I can ⁓ add into Mac OS for ⁓ for Telegram because I I built that endpoint and and ⁓ I have the telegram bot. Yeah, of course. Like that's Really easy. I mean like it doesn't it needs like two buttons. Yeah, no, like that is completely me overthinking and and and not like that that yours is the right answer. Having something quick, simple, personal owned. It's gonna be a short little blip of code that ⁓ that gives you a a custom endpoint to input stuff. So ⁓ so thanks for pointing out that sometimes we miss the the force for the trees. This is what I do, Veer. I can't turn it off. I can't. So you've got those surfaces. The surface for input that is inputting one like a thought, a task, an item, a memory, some kind of item for hey, database, store this for me so I don't have to think about it or have to, you know, hold that in my own personal memory.

12So there's d that surface. The other surface that I've built out are a set of right now, it's Python scripts, but again, it could be something where It could be a folder that is ⁓ always listening for information to be dropped into it. But there's scripts that will read a document, break it into what's called chunks, and then use the this ⁓ embedding layer to ⁓ stick it in the database, but also build the relationships between those chunks. And for me, an example would be is that I have an a consulting deck that I've used before that has ⁓ details for how to conduct a ⁓ an ERP an RFP process for for ERP ⁓ implementation vendors. ⁓ and it has rules and things to think about and you know, it it's it's just a it's a tool set for me. ⁓ when I do a consulting engagement where I'm helping somebody do a ⁓ an ERP project or to help them choose a vendor, I can drop that in and now that information becomes something that is searchable and reusable if once it gets put into that that memory layer. Yeah. Yeah, no, I think that's great.

13⁓ for me, I did ⁓ I've done quite a bit of pulling data in, ingestion. I have Manaus, which is a tool that I like to use. That has a really nice API. So that was fairly straightforward. All of that manus data is pulled in. I think I have it set up daily. comes in, it's working just fine. ⁓ and I their doc their API is really well documented, so that's no big deal. And then I did write Python scripts to get all my data from Claude, all my data from Chat GPT, and run through that. And then it doesn't duplicate any data. So there's logic in there that allows me to just because right now the way it works with Chat GPT, which is really frustrating, is is you download your data, wait three days or two days. I don't know, maybe it's quicker than that. But I waited some amount of time between one and three days. And then I had to take that gigantic file that has multiple files in it, right? It's a zip file with a ton of different files in it. And the only thing that I have ingested of Chat GPT right now is the text.

14So my conversations, there's a there's a file for conversations that my script runs through. Yeah. And pulls all of that into the database, processes it through open AI, and chunks it, so takes relevant chunks of data from that information. And then the way I have it set up is it hits like kind of a raw events table. And then that gets converted. The most relevant stuff, as determined by AI, gets written to a memory items. table. And at this point I probably have thousands of ⁓ more I have lots of raw events and then several thousand memory items that are being that are getting this semantic layer, this this easy way for a computer to find the memories. Takeaway here is that you are creating memory items. And that goes back to that conversation of the things that deserve to be remembered. That's what's happening here. Things that are getting stored in this database is not just raw text. It's decisions, thoughts, ⁓ epiphany, whatever. Things that like like that are getting stored away because that's where value is, not in.

15The sentence that got said during a ⁓ a chat or a meeting about where we're gonna go get lunch today. If we try to put a little bit of a bow on this, is we've talked about what it is, what are the meaningful pieces that go into building the machine, right? So if we were ⁓ if we were building a car, we've talked about the factory, we've talked about the pieces that went in. Right. But we've not driven the car. We've not shown you what and what the car is capable of doing yet. So ⁓ but but we've walked we've walked you through how f assembly of the vehicle works. and ⁓ maybe even how the paint shop goes on it. Give a little bit of what is to come in the next installment of this conversation is okay, now we've got this brain. What are the cool use cases that will really help us extract value? And I think you've done most of the work on that so far, Veer. So you're gonna have a lot to talk about with your system that you named Jarvis. Named it Cyrano. That's what I meant. That's what I meant.

16And you know, it's such a clever name, I should have remembered it. You said it like a million times too. No. It's a a ⁓ I I I'm a little proud of it. ⁓ it's a good name. ⁓ but ⁓ I I do think that the idea Go lunch with Mary. You haven't done that for two months. It's really it's clever. Exactly. I know. So we've got this product. We've talked about the fact that it is it's nerdy. It is something that is a little bit of a science project because even in the time from when I started doing some design work, ⁓ I've got several pieces that I'm still continuing to build and I'm redoing a couple of pieces of it because I want to add a little bit more detail here and there. It's it's an evolving work, but the value that we're trying to get out of this, and maybe we end with maybe a little ⁓ I'll do the very ⁓ hopefully a quick story, which is what we started with. So I'll I'll do the callback is we started with, you know, is AI going to, you know, take over the world? Is it dangerous?

17There was an article by ⁓ Thomas Friedman ⁓ earlier that the world the world Thomas Friedman. is flat. That's I mean that's the book that made him that's it that's like what really put him on the map. Right. And he was talking about some advances in AI that that one of the major frontier model ⁓ companies was talking about, ⁓ kind of ⁓ saying that it is dangerous and they're trying to limit its exposure right now because of the power behind it. And ⁓ I don't know if I buy into that, but ⁓ i i it it absolutely, even when Chat GPT first came on the scene and it became kind of this ⁓ internet darling, everybody who didn't quite understand what was happening, kept, you know, doing the Skynet conversation about like it's gonna take over and then the Terminators are coming for us. ⁓ this product that we're building, the goal, a a long-term goal of this is building a second brain, but a partner, something that can be agentic.

18⁓ you know, my one of my goals is not even that I would come in in the morning and ⁓ do a knowledge dump with it and have it help organize my day. I want it overnight to kind of wake up, look at my calendar for the day, identify what my highest value tasks that I ought to be working on. Look at my calendar, look at who I'm meeting with, prepare me kind of a briefing so that when I sit at my computer in the morning, I have some ideas of something I ought to be working on, who I'm meeting with, maybe Things I could be following up with because Jim mentioned, you know, ⁓ tinned fish and sardines. And I we're gonna talk about that when I talk to him next. And and I want it to s to have identified and maybe put a folder on my desktop that says these are files that you probably want to review today. One's maybe a dossier on ⁓ somebody you're talking with. Here are the two ⁓ presentations that you're already working on, or there's links that to to ⁓ deliverables that are being worked on. And I described that to my wife and she's like, that's creepy.

19And I'm like, well, I think we're moving to a world where the expectation of delivery is much higher than anybody is is maybe willing to say, especially for the type of work that is expected of of ⁓ consultants and of executives. And we're gonna maybe move to a world where There's not going to be a good excuse for not doing it. Yeah, one hundred percent agree with that. I think I've had a lot of conversations over the past week with different people across different industries that want to learn AI. And it's made it very apparent to me that there aren't a lot of folks that are willing to sit down and listen to conversations like this and as detailed as you think this is. These are these are things that I I truly believe you should get your mind around if you, you know, plan to be in the workforce and be very relevant. ⁓ and and maybe be outside of, you know, any trades or professions where you use your hands and you're doing some really physical things. If you're doing knowledge work, I think learning to live symbiotically with AI and the capabilities that it provides you.

20That is that is the way forward. I think so. So I think we've got more to talk about about, you know, now that, you know, maybe we've built the car, we can talk about what the car is capable of doing. What would he what are you know, how how do you drive this car? Because there's some rules. ⁓ but also you know, what what will it look like as we go forward? You know, we've laid out one use case or two use cases of where this is going. ⁓ but this is where Hundred percent. there's an opportunity for folks to be engaged. And again, this is right up at the front of the pack. ⁓ Jim and I talk about, and I, and this is, you know, we're not talking about getting bleeding edge. We're not building chips, but we're running at the front of the pack. And ⁓ that competitive advantage is not necessarily going to be there ⁓ in eighteen to twenty four months. I think that we've laid out one way that you can stay very competitive.

21You know, in the worst possible case, you've learned so much more at the end of your journey trying to build an open brain that can help you accomplish so much more. That that is it that's a skill that people will want. Just the knowledge of how to do it, how to build something like this is a huge leg up.