The argument in one line.
A knowledge base only works if it can find the single most relevant, most recent, best-sourced fact and inject it directly into the prompt — retrieval quality beats document count, and that quality comes from tagging every entry with source, timestamp, and authority before it's ever queried.
Read if. Skip if.
READ IF YOU ARE…
- You run a small team or solo business and keep losing institutional knowledge in scattered Slack threads, docs, and emails.
- You're comfortable directing an AI coding agent (Claude Code or similar) with plain-English instructions and letting it handle API setup.
- You want a working mental model of retrieval-augmented generation (RAG) before building or buying a knowledge-base tool.
- You're evaluating internal AI tooling and want a real engineering example instead of a vendor pitch.
SKIP IF…
- You've already built or shipped a production RAG pipeline — the RAG explainer here is introductory.
- You want a plug-and-play SaaS knowledge base with a UI; this is a build-it-yourself walkthrough that assumes agent-driven coding.
The full version, fast.
Cerebras published a blog post on the internal knowledge base its employees ask 15,000 questions a day, and unlike most "second brain" content, it's a genuinely practical retrieval-augmented-generation (RAG) system, not a visualization gimmick. RAG works by retrieving a specific fact from a database and prepending it to a prompt before the model answers, collapsing a vague question into a precise one. Cerebras's version ingests Slack, Wiki, code repos, and internal databases, converts each into an embedding tagged with metadata (source, timestamp, author), and weights retrieval by recency and authority so newer, higher-trust information wins. Nick rebuilds a scaled-down version live using Claude Code, connecting it to his own Slack, Gmail, GitHub, and YouTube, and shows the resulting system answering 17 of 20 real business questions correctly versus zero without it.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:33
01 · Cold open — not total bullshit
Nick names Cerebras directly and stakes the claim: unlike most "second brain" content, their blog post shows a real ingestion pipeline, not a graph visualization gimmick.
00:33 – 01:47
02 · What they actually built
Slack, Wiki/Confluence, GitHub, Netlist/PRM docs, and custom databases all get fed in and squashed into embedding space so anyone in the company can query what's happened.
01:47 – 02:25
03 · Why this isn't hype
This is one major company's practical engineering choice, not a creator's personal system — it's not built to look impressive to non-technical viewers.
02:25 – 03:53
04 · RAG 101 — the 'how tall is Nick' example
A vague prompt gets a vague answer because the model enumerates over all its data. RAG fixes this by retrieving a specific fact and prepending it to the prompt.
03:53 – 05:27
05 · Collapsing the variability
Prepending "Nick is six foot two" before the question collapses every possible model response into exactly one correct answer — that's the entire mechanism.
05:27 – 06:37
06 · Anatomy of a knowledge base
Three components: a platform to collect and store internal data, a platform to query it, and an auth/auditing layer (Nick skips the third for his own smaller build).
06:37 – 09:53
07 · Data sources
Slack, Wiki/Confluence, code repos, netlists/PRM docs, and custom databases are each scanned and converted into embeddings — any messaging or documentation platform works the same way.
09:53 – 12:29
08 · What an embedding actually is
Raw text gets passed through an LLM and tagged with metadata — time, subject, description — much like a photo carries EXIF data alongside the image itself.
12:29 – 14:24
09 · Recency and authority weighting
The system weights new information more heavily than old, and higher-authority sources (the founder) more than low-level ones, the same way human memory prioritizes recent, important events.
14:24 – 16:05
10 · Querying and the Slack-thread example
A support thread gets divided into windows and distilled by a small model into a structured artifact: question, summary, resolution, systems involved — permanently answerable going forward.
16:05 – 16:45
11 · You don't need a dev to build this
Nick pivots from explaining the mechanics to building it himself with a coding agent — you don't need to understand every internal detail, just the data flow.
16:45 – 18:39
12 · Live build — prompting Claude Code
Nick pastes the Cerebras blog post into Claude Code and asks it to build the same system for his own business, connecting Slack, Gmail, GitHub, and YouTube.
18:39 – 20:00
13 · Setting up access from scratch
For an account the agent doesn't already have access to, it explains the exact setup steps — Google Cloud project, Gmail API, OAuth consent screen, refresh token — roughly 5–10 minutes per source.
20:00 – 20:40
14 · Visualizing the build in ASCII
While the build runs, a second Claude Code thread renders the system architecture as an ASCII diagram: connectors, distillation step (via Haiku), one Postgres table, full-text plus embedding retrieval.
20:40 – 21:14
15 · The scoreboard: 17/20 vs 0/20
The same model is asked the same 20 real questions with and without the knowledge base connected — 17 correct with it, zero without it.
21:14 – 22:08
16 · Live query examples
Nick asks the system which company recently pitched sponsorship for his podcast (Abacus.ai) and which GitHub repos he pushed to most recently — both answered instantly and automatically, no manual lookup.
22:08 – 23:48
17 · Why this beats naive RAG, and the close
Retrieval quality, date/time weighting, and source authority matter more than raw document count — Cerebras has hundreds of thousands of documents, Nick has 600, and both work because of ranking, not volume. Closes with a low-pressure Maker School plug.
Lines worth screenshotting.
- RAG is just retrieval augmented generation: pulling a specific fact from a database and inserting it into the prompt before the model has to answer.
- A vague question like "how tall is Nick" only gets a vague answer because the model is enumerating over all its training data with no specific context.
- Prepending one line of fact — "Nick is six foot two" — collapses every possible model response down to exactly one correct answer.
- A knowledge base is only three things: a platform to collect and store internal data, a platform to query that data, and an authentication/auditing layer.
- Embeddings aren't just raw text stored as-is; each piece of data is passed back through an LLM and tagged with metadata like timestamp, subject, and description before it's stored.
- The system weights information the way memory works: data from a founder thirty seconds ago outranks the same topic from a low-level employee three years ago.
- Retrieval quality, not document volume, is what separates a good knowledge base from a bad one — Cerebras has hundreds of thousands of documents, this rebuilt version has 600, and both can answer well if retrieval ranks the right ones first.
- Cerebras's own employees ask its internal knowledge base roughly 15,000 questions a day, three months after launch.
- In a head-to-head test, the same model answered 17 of 20 real business questions correctly with the knowledge base connected versus zero without it.
- You do not need a developer to build one of these: an AI coding agent can scaffold ingestion pipelines for Slack, Gmail, GitHub, and other sources from a single plain-English prompt.
- When a coding agent doesn't already have access to a service, it can walk you through exactly how to get it — Google Cloud project, OAuth consent screen, refresh token — in about five to ten minutes per source.
- Every data source, from Slack threads to code repos, lands in one single embeddings table, so the query interface stays identical no matter where the underlying data came from.
Retrieval quality beats document volume, every time.
WHAT TO LEARN
A knowledge base only works if it can find the single most relevant, most recent, best-sourced fact and hand it to the model before the question is even answered — and that's a data-tagging problem, not a document-count problem.
04RAG 101 — the 'how tall is Nick' example
- A vague question gets a vague answer because the model is enumerating over all its training data with zero specific context to narrow it down.
05Collapsing the variability
- Prepending one specific fact to a prompt collapses every possible answer into exactly one correct answer — that's the entire mechanism behind RAG.
06Anatomy of a knowledge base
- A knowledge base is only three things at its core: a way to collect and store data, a way to query it, and (for larger teams) an auth/auditing layer.
08What an embedding actually is
- Raw text alone isn't enough — tag every stored entry with source, timestamp, and author so the system can rank results instead of just matching topics.
09Recency and authority weighting
- Weight retrieval by recency and source authority, the same way memory works: a fact from the founder thirty seconds ago should outrank a stale note from years back.
10Querying and the Slack-thread example
- Every source, from chat threads to code repos, should land in one unified table so the query interface never has to know where the data originally came from.
12Live build — prompting Claude Code
- You don't need to understand every implementation detail to build one of these — an AI coding agent can scaffold real ingestion pipelines from a plain-English description.
13Setting up access from scratch
- When a coding agent lacks access to a service, ask it how it would set that access up from scratch — it can walk through OAuth, API keys, and tokens directly.
15The scoreboard: 17/20 vs 0/20
- Test any retrieval system head-to-head, same questions with and without it, to prove the return is real before rolling it out to a team.
17Why this beats naive RAG, and the close
- The differentiator between a good and bad knowledge base isn't total document count — it's whether the highest-quality, most relevant results actually surface first.
Terms worth knowing.
- RAG (Retrieval-Augmented Generation)
- A technique where a system retrieves a specific, relevant piece of information from a database and inserts it into a prompt before the AI model generates its answer, instead of relying on the model's general training data.
- Embedding
- A numeric representation of a piece of text or data that lets an AI model compare and retrieve it by meaning rather than exact keyword match.
- Embedding space
- The searchable collection of all stored embeddings that a retrieval system pulls from when answering a query.
- Distillation (in this context)
- The step of passing a raw conversation or document through a smaller, fast language model to extract a structured summary — question, resolution, participants — before storing it.
- MCP (Model Context Protocol)
- A standard that lets AI agents connect to and query external tools and data sources, such as a company's internal knowledge base.
- Recency weighting
- Ranking retrieved information partly by how recently it was created, so newer facts are preferred over older ones on the same topic.
Things they pointed at.
15:56toolClaude Code
20:34toolAbacus.ai
23:40productMaker School
Lines you could clip.
00:00
“This company, Cerebras, just built a knowledge base that I think is not total bullshit.”
11:45
“Much like your brain remembers new things more, better, and stronger, this system remembers new additions to the knowledge base more, better, and stronger.”
22:13
“The thing that differentiates a crappy knowledge based system from the ones that are good isn't necessarily how many documents you have, but it's how effectively can you retrieve the highest quality ones.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00So this is really interesting. This company, Cerebras, just built a knowledge base that I think is not total bullshit. And that's hard for me to say because up until now, virtually every instance of knowledge bases or second brains or whatever have just been total hot air.
00:17This is a massive AI hardware company that essentially creates super quick inference chips, uh, and they're used by a variety of major billion dollar companies that you guys probably have heard of. And at no point in this big tutorial, which they just wrote a big blog post on, did they show an Obsidian graph. At no point did they show a .
00:37Uh, you know, instead, what they showed is they showed a very reliable and robust data ingestion pipeline that allows virtually anybody in their company to answer any question that they have about the business and about things that other people have done in the business. What they've done is they've fed in, like, gigabytes of text and image data from platforms like Slack, every Slack conversation that's ever occurred in the organization.
01:00Um, their Wiki and Confluence, all of their code repos in GitHub, you know, their Netlist, PRM docs, as well as custom databases.
01:08They've just slashed, or rather I should say, squashed all of this data into embedding space, which you can just think of as concepts that an AI could query over.
01:17So that if somebody said, hey, what was that thing Paul was talking about three days ago? It had something to do with, you know, embeddings. It can actually go in to the Slack without having to, you know, query the API or whatever, and just always know what's going on in the business.
01:30So what I wanna do in this video, but isn't clear, is I wanna show you what this is. I'm gonna actually wanna build one alongside here. I've already built one of these for my own business, and it's quite useful already.
01:39And that's just as a small team. I can imagine how if you have thousands of people in your organization, stuff like this starts to scale and get really, really good. Um, and this is also just, as mentioned, like one of the ones that's not bullshit.
01:51There's a lot of noise on YouTube, obviously, because everybody's trying to push their own particular way of doing things. This is not my way of doing things.
01:57This is this major company's way of doing things, um, that I happen to think is actually quite practical in it. It is not all about visualizing something goofy for people that are not super technical. Now, to really understand this system, you first need to understand what retrieval augmented generation is, also commonly referred to as REG.
02:14RAG is the cornerstone behind more or less all of this stuff. So it's worth at least a brief refresh if this is the first time that you're encountering this or not. What RAG basically does is it allows models to answer questions specifically about a set of data, as opposed to just in the general sense, overall data.
02:31Like, instance, if I were to type how tall is Nick into my AI model chat box right now, it would probably ask me, who are you talking about? Who is Nick?
02:40Well, that depends broadly on which Nick you're talking about. If you meant, you know, Nick Jonas. Nick Jonas is x y z tall.
02:47Right? That's a very general question, so it's likely that you're gonna have a general answer. And the reason why is because it's enumerating over all of its data.
02:53So just to visualize this and because I thought it'd be kind of funny to make the whole world know I'm taller than Nick Jonas. You know, here's a quick visualization of maybe all the possible answers you could get. You know, how tall is Nick?
03:02Which Nick? Nick Jonas is five foot ten. That depends.
03:05Right? Now, the question is, imagine, hypothetically, if I didn't just say how tall is Nick.
03:11Imagine if the prompt that I fed into this model was Nick is six foot two. And then I added some lines over here.
03:20And then I said, how tall is Nick? What do you think the model would say? Well, it's pretty obvious how tall Nick is.
03:25Right? We just fed in a piece of information immediately before the rest of the prompt that just says, how tall is Nick? So what do you think the model's gonna say?
03:32Well, the model's probably just gonna say, well, Nick is six foot two. It doesn't even know really which Nick I'm referring to or anything like that because it it just has the information directly in its prompt. Right?
03:43And so what we've done is we basically collapsed all of the possible responses, all the variability of the model, and made it hyper specific. We basically made it like exactly what we want every time. Okay.
03:53So all RAG really is, retrieval augmented generation, is it is augmenting the AI generation of a piece of text by retrieving information like Nick is six foot two from some database.
04:06K. Now the key is, it's not just a database. It's not like we just hypothetically have heights of every man on planet Earth, and then somehow we store the right entry in.
04:15What it usually is is it it sort of exists in a language that LLM and large language models can intuitively understand called the embedding space. You should know that all retrieval augmented generation really is is it's just a system that allows us to store information before the question. Now, hypothetically, let's say you were trying to build a really cool knowledge base that just answered all the questions inherent in your business, everything that's gone on in the last twenty four hours.
04:38Um, well, you can apply that same approach. Like, if I just asked my little knowledge base, hey, what did Peter say at that talk last week? It's probably gonna go and collapse into a or rather spread a variety of different answers.
04:49Gonna be like, who the hell is Peter? What talk are you talking about? I don't even know what year it is, man.
04:53I'm just a a text corpus. So what we need to do is we need to find a way to insert information. So, you know, realistically, what we want is like Peter Jackson delivered a talk last week at 06:52PM on Thursday.
05:07In the talk, he discussed x y and z. What we want is we basically just want a way to prepend that before the prompts. Right?
05:15So all this knowledge based system from Cerebras does is it just allows you to basically retrieve a relevant piece of information from your company and then inject it above your prompt. Such that if this is your main prompt over here, this is just like your sort of injected, and I don't know.
05:32I'm just gonna call it like help. I mean, the model doesn't know the difference. Right?
05:36The model just sees the text. Whether it saw this or whether it saw all of this combined, you know, it's gonna heavily change what the model says. But it it doesn't know that, like, you know, we've helped it out by providing that information.
05:46It just thinks like this is part of its training data, essentially. Um, and so that's that's more or less what the system does. And that's why it's so powerful.
05:52It allows anybody in the whole company to answer any question about anything that's gone on. It also allows you to store things like, hey, you know, here's a set of problems, and then here's a set of solutions that we use to solve those problems. So if a newbie comes on board, they're, I don't know, trying to make some code fix or something, and it's not entirely working.
06:07They can just go to the knowledge base and say, hey, like, why isn't this thing working? And then it would say, ah, you're probably not pushing to prod as we do in this business. It's a little bit different than most other businesses.
06:15Let me show you guys how it works. Alright. So the first thing to know is this is made by three handsome gentlemen, Isaac, Daniel, and Mike.
06:21At least they're the ones that wrote the the blog post. And currently, they're being asked something like 15,000 questions every day. Now, this is obviously a pretty long blog post.
06:29They're a bunch of really cool visualizations and stuff, but for the sake of your time, let me find the most important sections of this, so that you have everything you need. Okay. The first is the anatomy of the knowledge base.
06:39And so this knowledge base is composed of three major sections. The first is it's a platform for collecting and storing internal data.
06:47And so that makes sense. Right? We need to find a way to basically get knowledge from outside of, you know, the system into the system.
06:54So the knowledge inherent in conversations, let's say, of people in the Slack. The knowledge inherent in the SOPs of the company, the the checklists, the processes, all of the emails, all of that stuff.
07:05We basically need a way to take all of this in the ether and then stick it into this knowledge base in an appropriate manner so that it doesn't get super crappy. The second thing we need is we need a platform for querying that data. And so now that we have the data in the the brain, so to speak, the knowledge base, we need a way to extract it and kind of pull it out.
07:21And so this, for instance, would be if I asked my knowledge base a question about, hey, grab stats on the last week of performance for x y z software platform, and then apply that cool lens that Nick talked about in his most recent company seminar seminar to to it. It.
07:37That would be an example of, you know, using this platform which we've collected and stored internal data into to query it and then pull something out. And then finally, because Cerberus is, you know, a big company and they're dealing in hardware, they also need a layer that enforces authentication and authorization with auditing, analytics, see who's like accessing the data, why, and so on and so forth.
07:56Now, I'm pretty small relative to this massive hardware company that works with billion dollar businesses all over the world, so I'm not gonna do that. And really, the thing that's most important for us is just gonna be one and two. And I'm gonna assume anybody in my business is probably okay.
08:08So let's actually visualize what this looks like. Basically, at the very top end, we have a bunch of different data sources that they're constantly ingesting, right, for that first section to to to store a bunch of data in their platform. And the first is Slack.
08:23So as I'm sure you can imagine, any major business will have some sort of messaging platform these days. And basically, what this does is it scans through all of the messages that are sent in all of the accessible workspaces, and then converts them into what are called embeddings.
08:36And embeddings are just another way of referencing or referring to that data. It's just it's a little easier for models and LLMs to understand. And so you can imagine, you know, if you guys use Discord, you do this on Discord.
08:46If you guys use email, then you wouldn't do Slack, you just do it over email. But the idea is you just need a place where people are talking. The second is, you know, most companies, at least sizable businesses, will have some form of Wiki.
08:57So whether it's Wiki, whether it's Confluence, whatever it is, um, there's some place that you guys have already tried to sort of store all of your information. And typically, that's done naively through, like, checklists and SOPs and onboarding pages and welcome to Clarvo. You know, this company is gonna walk you through blah blah blah blah.
09:13And so just like Slack, you sort of stick that into this embedding space. Next up are code repos. This is particularly important if you're a software business or software adjacent.
09:21And in my case, I consider my business software or software adjacent. So, you know, scanning through all the code repos is gonna be important. Net lists, PRM docs, you know, this stuff is gonna be more, uh, relevant again to software businesses.
09:32And then also any sort of custom databases you have that store information. And so maybe you are doing some sort of sales or rather spearheading a a sales department, I should say, and you guys have databases that have info on all of your KPIs. So, you know, how many sales are being made, how often.
09:46You know, what what we can do is we can actually take all of that data, that formatted database information, and then just store it as these embeddings. Okay. And that sort of takes me to that idea of like what is an embedding.
09:56Well, an embedding basically just takes the raw data itself. So let's say hypothetically, the data that we wanna stick in this pipeline is just a I don't know.
10:06It's it's Nick thinks, I don't know, penguins are cool.
10:11So what we have is we have this naive sort of text data here. And then what we'll do is we'll actually pass that through a large language model. So this might be, I don't know, man, like ChatGPT, Claude.
10:20It could be an open source model. It could be anything. And then we just ask it questions about this.
10:24And we actually say, hey, who's Nick? Why did Nick say that? What time did Nick say that?
10:28Where did Nick say that? When did Nick say that? And so on and so forth.
10:32And so, at the end of it, we don't just have this naive piece of text that says Nick thinks penguins are cool. We have Nick thinks penguins are cool. You know, time, 05:32PM.
10:42We have, I don't know, subject. Nick Sarayev. We have description, you know.
10:47And then here, maybe you would say something like, you know, oh, Nick is the founder of x y z organization. Nick does blah blah blah and blah blah. So what we're doing is we're basically taking a core piece of data, which you can visualize sort of like, uh, I don't know, just this cube over here.
11:00And then we're adding metadata to it. We're adding a bunch of additional kind of outcrop pieces of information.
11:05So it's no longer just that tiny little nugget in the middle. Now it's like, okay. Now we know the time of data it was sent.
11:10We know who it was sent to, where it was sent from, who the person is, why any of this is relevant, what the point is that Nick was trying to make, and so on and so forth. And, I mean, it's pretty similar to, I don't know, like, way that a camera works if you think about it. When you take a photo with a digital camera, right, you have the photo, which is the core data, but then you also have a bunch of metadata around that.
11:28You have stuff like, you know, what time it was taken, what the color scheme of the photo was, where, you know, a lot of these cameras are now storing GPS info, which is kind of terrifying. But anyway, what I'm trying to say is we're not just storing the data itself. K?
11:40What we're doing is we're adding a bunch of stuff to that data and then storing it in a way that the machine understands. And the reason why that's important is because much like your brain remembers new things more, better, and stronger, This system remembers new additions to the knowledge base more, better, and stronger.
11:57And so we have the source, we have the timestamps, we actually know how to weight the data. The data comes from some super low level employee three years ago. We're gonna weight it a lot less strongly than if the data comes from Nick, the founder of the business, and it happened thirty seconds ago.
12:10You know, somebody asks for information related to some recent seminar or something like that, it will go through not just all seminars equally, like most naive language models would do. It'll go through the most recent seminar, uh, sort of preferentially. It'll be like, okay, recency.
12:23What does that mean? Okay. What's the date?
12:24Okay. Cool. And then in this way, you're gonna have a bunch of these, like, quote unquote memories of this brain, um, that are sort of, uh, ranked based off the date, based off the time, and based off the relevance.
12:34Okay. So that's that. And then the third step is obviously actually querying that data.
12:39And so that data is queried, you know, via MCP, you know, WebUI, AI agents, people are are sending messages into a chat box, whatever it is. You know, what what occurs really is it goes in reverse.
12:50The query goes through the embeddings. The embeddings represent a particular form of information, say a Slack message.
12:56K? And then, you know, we what we do is we pull the embedding version of that in order to get you the answer. Key practical example of this with a Slack message, and, you know, I think the examples I provided here do kinda suck, so bear with me here.
13:09But there'll be a message maybe sent in the CKPT dash support thread. K? And at 09:14, Maya will say, hey, the restore stalls after manifest load on the larger cluster, but small runs are fine.
13:20Maybe Owen then says, yeah, can reproduce this with 128 chars. The logs stop before the cache warm up though. Basically, what happens is question summary.
13:28Right? Maybe here's a resolution a few minutes later. We take all of the information in a chat thread, sort of divide it into these windows, and it'll all be done for you automatically.
13:37And then it'll be essentially created as a structured artifact. She just can't really see her, so let me zoom in. And so this is the metadata that I was talking about.
13:45Because we'll have a large language model, um, take this information and ask questions about it, you now have things like question. Hey, why does the restore stall after the manifest load? Well, guess what?
13:54The summary of the sort of this conversation is, well, large restore stop before the cache warm up. You know, if I knew what the hell this meant, it might be a little bit easier. But maybe it's like, hey.
14:03Who was that guy that came in to do that talk on Thursday? And maybe the summary is now, you know, the guy that came in to do the talk on Thursday was Nick.
14:11And maybe there's a resolution step if it's something that needed to be resolved. And so in this case, you can imagine how now we have the ability to actually, I don't know, like, literally store the answer to this question permanently inside of the company, meaning the company grows more efficient and more effective over time.
14:24You then store a bunch of things like maybe code rest. Maybe that just means, I don't know, like what is the Slack thread where all that information is stored. Maybe systems that includes a bunch of systems that these people are working on.
14:35It'll actually store a list of them. Obviously, it includes other information as well. It includes, like, the source ID.
14:39It'll include, um, I don't know, the time and and and date and so on and so forth. But yeah. I mean, like, that is that's about how detailed we're getting under the hood.
14:46Uh, but we, again, don't need to store any of that stuff. Just wanted to make sure you guys all understood what was really going on under the hood. So you don't need to know everything that I just said.
14:54That's okay. I'm gonna walk you guys through the actual build process. And the good news is I don't even know, you know, what 90% of this stuff really is.
15:00I just understand it at a high level, which is sufficient. As long as you understand the way that the data flows, we can actually take that and use that to build our knowledge base. You don't need a crazy dev to do this.
15:08You know, we can a 100% do this ourselves. Okay. So how do I actually build something like this in practice?
15:12Well, the very first thing you need to do is you need to whip up a coding agent of some kind. And so in my case, I'm using Claude code v two point one point two one one. I'm not saying you have to use Claude code.
15:22You can use one of many available open models, but this Fable five intelligence right now just happens to be the best, aka the most performant, and I have the money to to pay for it, so I might as well. Okay. So you can get whatever model you want, Claude code, you know, ChatGPT codex, whatnot.
15:38As long as it's like a coding model, it can actually make changes in your computer, you're good. If you want more information on that, then just, you know, check out my my four hour Claude code course. So all we really need to do to make this work is I'm just gonna copy this whole blog post in.
15:50K. And I'm gonna go all the way back to my ghost TTY terminal, that's just what I happen to be using for this. Then I'm gonna paste this in.
15:57Then I'll say, I'd like you to build something like this for, and in my case, I'll say Nick Sorayaev Media LLC. I have a Slack, a set of emails, an email address, a GitHub, and a YouTube channel.
16:12I want us to build ingestion pipelines, which is just how we get data into the system for all of them. I'll also say, PS, this is a demo. I've already done this.
16:22Given that's a demo in front of the audience, I want you to build it totally fresh. The reason for that is because I've just already done this several times, and what I wanted to do is just pretend that it's totally new. Now the cool thing is, as mentioned, you do not need to know how this works.
16:35You can just go through this, k, entirely autonomously. Uh, the model can sort of figure it out. Um, if for whatever reason, Cerberus has, I don't know, put a block on their page, then rather than feeding the URL, you can also just go back to Google Chrome, copy the whole thing manually, and literally just paste it in.
16:50Um, the important thing is you just need to get it in to, like, this intelligent model. Now what it'll do next is it's going to attempt to log in to all of the various accounts of reference. So in my case, I actually have accounts that are already logged in.
17:02For instance, I have GitHub logged in. I have my YouTube channel logged in. I have my email addresses logged in.
17:09And so it'll do so almost entirely automatically, basically, at once. But I want you guys to know that that might not necessarily be the case for you.
17:16And so if you guys wanted, for instance, an email to be set up, all you have to do is just open up a new instance of this. I'm just gonna go claw dash dash for both. So I'm just gonna zoom in a little bit so you guys could see both of these.
17:27And then I'll say something like, hey, I want you to get access to my email. Then I'm just gonna give it one of my emails, nick@leftclick.ai. You should be able to query each email, read all of my emails, send emails, basically have full admin access.
17:45How would you do this hypothetically if you didn't already have access to it? Now, I'm asking that question because it already has access to my emails and I don't wanna sit here for four hours showing you guys It say, I already have access to your email.
17:59And as you can see, it eventually will tell you, here's how you'd set it up from scratch. You would have to open up a Google Cloud project, enable the Gmail API, do an OAuth consent screen to create an OAuth client, some sort of desktop app.
18:12You would have to do a one time consent form. You would also need some sort of refresh token. I guess what I'm trying to say is it's not it's not very difficult.
18:19This would realistically take you about five to ten minutes for every pipeline of information that you wanna set up. And so that's the cool thing about agents. I mean, you can just ask them, hey.
18:27How do you do this? And if the model is smart enough, in our case, uh, Opus 4.8 in this instance, but, you know, you could use Fable five. You could use GPT 5.6.
18:34So you can do whatever you want. Um, you'll get the credentials that you need in order to build set ingestion pipeline. And the left hand side, speaking of, it is currently doing the building.
18:43Now because I kinda want to see what's going on, I'm actually just going to go slash b t w. That's a Cloud Code specific feature. And I can say, visualize the system you're building for me in ASCII.
18:54In case you guys didn't know, ASCII is just a text way to demonstrate sort of visually like what a diagram like a diagram or something like that. And so what this is now doing is while the main thread is running, I have another thread that's going on on top of that, just asking it various questions. And so I'm saying visualize systems you're building and asking.
19:11Here's what it's doing. It's actually showing me it. So we could see.
19:13I'm just gonna zoom out. And so what we're doing is we're basically building that exact so we have a Slack over here. We have three accounts over Gmail.
19:20We have my GitHub, and then we also have my YouTube channel. And so these are the connectors. So this is sort of like the APIs and, uh, MCPs and whatever it is that allows us to pump stuff in there.
19:29We have the distillation step, like I talked about earlier, where we're basically taking each thread, passing it through a Haiku in this case, feeding in a question, a summary resolution systems, And then we're ingesting this into our actual pipeline or database for giving it just one table in what looks like PostgresQL.
19:45And then down at the very bottom here, we have retrieval setup with full text where we look for exact matching tokens and stuff like that and then embedding. Okay. And right after this is all done, I'm just gonna jump to it, and then I'll show you guys how it actually works when you query it.
19:58And now you can see sort of a artifact that shows the same model asked the same 20 questions, one with the knowledge base and one without. You could see that the one that had the knowledge base answered 17 out of 20 questions correctly.
20:11The other three were just honest partials or declines. It just said, hey. I don't really know what's going on because I don't have the evidence to to back it up.
20:17Whereas without the knowledge base, I answered zero. So, I mean, these are really high quality comparatively. You can imagine how rolling this out inside of any companies would significantly improve both the quality and then the speed with which you get things answered.
20:29So for instance, which AI company recently pitched a paid sponsorship for the Stacked podcast? In case you guys didn't know, I run a podcast with Jack Roberts called Stacked. And, you know, we're we we've recently grown a little bit.
20:41I think we're up to 5,600 subscribers now. So Jack and I kinda screwing around having a fun time.
20:46We're almost at six k. Well, you know, if you just ask that question natively, obviously, you wouldn't get the answer.
20:52But in our case, we know it's abacus.ai. They pitched a paid sponsor for the stacked podcast promoting as chat LLM all in one AI super assistant. The average came from insert person here.
21:02Did we take it? Absolutely not. Or how about this one?
21:04Which GitHub repos did NSM push to most recently? We actually have the exact GitHub activity right over here, and it was just me replacing sort of a defunct company name with a new company name. Now this isn't super impressive.
21:17Any sort of rank could do this. What is impressive is the fact that this data is now being ingested completely automatically, and I have nothing that I need to do with it. It occurs entirely without my control.
21:26Not without my control, but not requiring my management. This will occur in the background. And so I can ask this any sort of question that I want at any point in time.
21:34Unfortunately, the thing you can't really visualize is just how much better this is than naive rag, which most other people are doing. Because we have information like the date and time, a k, how long it's been since the thing was created.
21:48We also have significantly higher quality REG. You know, I only have 600 documents inside of my system.
21:54A company like Cerberus probably has, you know, hundreds of thousands. The thing that differentiates, like, a crappy knowledge based system that you ask questions to and have woven into your business with the ones that are good isn't necessarily how many documents you have, but it's how effectively can you retrieve the highest quality ones, the most relevant ones.
22:10And that's the question that this sort of rag system answers. So I'm not gonna tell you I have to use this one. There are obviously a bunch of different rags that you can use, and there are also a bunch of different approaches to knowledge bases, but this is done by a major business.
22:21These guys aren't screwing around. They're not going to do something like this if it doesn't actually generate a return on their time and energy. And as it stands right now, this is definitely generating a return on time and energy.
22:31It's generating a return on time and energy for me because I have the syndicating and then embedding and then pulling and and enriching all content that I'm currently creating and that is going into my business. It's it's a it's an abundant training resource that immediately gets everybody up to speed in my team. Uh, you know, it's something that I can just dole out literally with the snap of a finger, and I can give them the same brain that I personally have spent a long time creating.
22:54And the unfortunate thing is, it doesn't look super sexy. I don't have a big floating brain in the middle of my screen, but, you know, it's the same resource that's accessible via Obsidian or via, you know, Graphify or whatever the heck. This is just probably the higher ROI way of doing it because of the frequency waiting, the time waiting, and stuff like that.
23:10Anyway, as you guys could tell, I'm simping hard for Cerberus. They're a dope company, and this is not a sponsored post in any way, shape, or form. I was not paid any money to do this.
23:19I just think it's probably the coolest knowledge based implementation that I've ever seen. So, yeah, that's that. If you guys like this sort of thing, definitely check out Maker School.
23:25It's my day by day accountability road map where I show you guys how to acquire your very first customer for an AI or automation service, which could be building rag or knowledge based systems like this for money within ninety days or you get all of your money back. So I have a refund guarantee because it works really well, and, obviously, I would love to see you there.
23:43Have a lovely rest of the day. Subscribe to the channel if you haven't already, and I'll catch all y'all on the next one.
The Hook
The bait, then the rug-pull.
Nick opens by naming the target directly: Cerebras, a chip company most people have never heard of as a knowledge-management authority, just published exactly how it built the internal tool its own employees now ask 15,000 questions a day. No Obsidian graph, no floating 3D brain — just a data pipeline he thinks is worth actually copying.
Named ideas worth stealing.
03:53concept
RAG (Retrieval Augmented Generation)
Retrieve a specific fact from a database and prepend it to the prompt before the model answers, collapsing a vague question into a precise one.
Steal forany internal tool, chatbot, or support system that needs to answer questions about a specific business instead of general knowledge
06:37list
Anatomy of a Knowledge Base
- A platform for collecting and storing internal data
- A platform for querying that data
- An authentication/authorization layer with auditing and analytics
The three functional layers Cerebras identifies as the whole system — most small builds can skip the third.
Steal forscoping an internal knowledge-base build without over-engineering it
12:29concept
Recency + Authority Weighting
Rank retrieved results by how recent and how authoritative the source is, not just topical similarity — the same fact from the founder thirty seconds ago outranks a low-level employee's note from three years ago.
Steal forany search or retrieval system where stale or low-trust results currently drown out the right answer
How they asked for the click.
VERBAL ASK
23:40product
“if you guys like this sort of thing, definitely check out Maker School... I have a refund guarantee”
single, low-pressure plug in the final 30 seconds, framed around the accountability program rather than the video's specific topic
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
