Introduction: Building Context and Ways of Working

absolutely phenomenal part of the things to understand is also is to build context let me just share my screen i'm trying to think where do i share share my screen uh screen share there we go so um let me just see window that one so all i'm going to do is i'm not going to show you how to write software all i'm going to show you is some of the ways that that i work on a day-to-day basis and hopefully some of that will rub up into you guys to figure out how to use these tools the most effectively most effectively possible

Create Your Own Knowledge Base

One of the things to remember, and I think a couple of the other guys already said it, is for you to collect as much information as possible and data that you should store yourselves and have that as an available database for yourselves.

So as an example, let me just show you our drive. We have a generative AI folder, and this is our Google Drive.

And we have a generative AR folder. And in here, there's a whole suite of research.

So some of you are techies, might understand.

Collecting Research and Papers

So for example, the techie guys there, the data scientist guys that are in the audience, one of the things we're doing is we're collecting all the papers around software engineering and how to build systems. And there are hundreds and hundreds of papers. You can see here there's hundreds of papers.

This is around how to use agents to build software. And there's research papers are out there. These are, as you can see here, this is just a sample.

Why a Personal Database Matters

so what we've been doing over the last three or four years is and the actual database that the sort of data document store here goes back 20 years so we've got 20 years worth of content but the thing to remember is you need you know one of the things to do with everybody is to build your own unique database of content and you can get that from everywhere you can download it you know this stuff is downloaded as you can see and and this is your knowledge base this is the stuff you know it's an extension of your brain that you can have, which can make you look amazingly intelligent and amazingly great at what you do as a job.

So whatever topic, whatever business line, whatever industry you're in, build your own database of content. And that will help you do your job. And actually, obviously, you can do other things with it, but be able to do your work way more effectively.

And that's what I do.

So what I've been doing is I collect huge amounts of research around the areas that we're working in and widen that in different industries, in different geographies, in different types of research. because it helps to build your own knowledge base, which then helps you do your job better.

Using Notion as a Central Hub

So one of the tools we're using at the moment is Notion. Hopefully some, most of you have heard of Notion and use it.

If you look in here, what I've got is, you know, this is just the AI bit, but the thing to look in here, as you can see here, these are all the different data sources you can connect to. So this is connected to our Google Drive.

It's connected to isn't connected to my email, but it's connected to other pieces. And you can add loads of different data sources in.

The most important thing is that now all of that data and data inside Notion.

Connecting Data Sources and Organizing Research

So one of the things that we have is going back to what I showed you in the research is the whole suite of paper research. So this is this is research that's out there going back.

three or four years so and all this is the ai one so you can see there's a lot of different a lot of different stuff around ai very specifically so you can see here this is if you go those of you people are techies you can see this diagrams looking at infrastructure and giving you ideas of how to agent k i for example right um And there's hundreds of papers.

And most of them are tagged. Some of them aren't because I forget to do it. The important thing is this is my knowledge base of all knowledge that I use every single day on what I do.

And then you can.

One of the things that we're doing is building tools for enterprises across different parts of their business. So the chaps just before talking about accounting.

From Research to Writing and Requirements

if you're an accountant and you work in accounting space that mcp stuff is great right but that's just an isolation bringing all the different content together will help you do lots more other stuff but it's great to start it's great to just get going because it helps you build your knowledge base and it helps you build that that extension of your brain that can help you do do things and then one of the things that i do on a regular basis is write papers so as an example i use perplexity

Sometimes I bring in some of that content that you've seen because you're in perplexity. If you ever use perplexity, you can actually bring your external content in.

And this is just an example of one of the things that we're building at the moment, which is a Gen BI platform. So how do you do reporting and how do you how do you move away from using Power BI or tablet? If any of you guys have heard of these tools and there's lots of other small tools.

to make doing business intelligence reporting a lot more dynamic a lot more hyper personalized for any customer so what i've done here is this is using some of our internal content but also going out and using this is using um deep research from complexity and it's written me a basically a a very detailed requirements document as you can see here let me just let me show you here you can see it's written somewhere here there's a csv file one second let's see if i can find it um so this is basically building our requirements how to build a platform and what we do is we take that and push it downstream into our dev cycle to build the epics and stories etc to build that platform right so

Prompting, Iteration, and Architecture Mindset

Effectively, if you had to write this by hand, it will take three to six months to do it.

And you can see, look at the prompt here. Can you see the prompt that I've written? And this is just the first iteration of the prompt.

Now, because I'm a techie and I'm an architect, I understand how to do that. I understand what I want in terms of architecture.

One of the key things and skills to learn is even if you're a software engineer or a data scientist, one of the things to learn, one of the things, the most powerful things you can do is understand how to build the structure of something, whether it's document or whether it's a piece of software, and design the system in your head and use that to construct prompts which will then get you frankly a lot better output and then you have to iterate through it so this is not one time you iterate through it and keep iterating to get um to get a way better answer and then what that does is allow you to effectively produce stuff amazingly quickly um and output that and use that for whatever you want so

Key Takeaways: Knowledge Base + Prompting

So to reiterate, one of the key things to do is learn how to build your own extension of your brain and all your documents. Secondly, understand how to do prompting and use prompting.

FAQs and Tool Comparisons

No, Notion isn't like Trello. It's a document store.

a doc it basically stores your documents and lets you write documents right right um it has a uh um somebody's asking about brett is asking about the notion it does trailer type stuff so yes you can build trailer boards and scrum and all of those things but notion is an amazing tool an amazing tool um and then one of the other chats demonstrated notebook lm so just to add more stuff to it

Notebook LM and Context Building

Here's an example where what I'm trying to do is these are some of the documents. I showed you some of the documents on our Google Drive. What I've done is built what's called a context.

And this context here is all about how to do software engineering with agents? And how do you do that?

And so what I've done here is I've been basically selected a bunch of our documents. And these are, you know, a bunch of the documents that we have around coding. And that allows you then to build in the context of that allows you to write a business requirements document.

So if you don't know, don't worry about what rag is this, this is system generation. So what we call what we're building, which is vibe you know about vibe coding we're doing what we call vibe engineering which is how do you build systems end to end and this is an example of explaining how that works and and being able to

From Context to Requirements and Mind Maps

take all the researchers out there and create a set of requirements that you can then feed downstream. One of the things that the chats didn't show a mind map, one of the things to look at is this kind of stuff, which is, I like notebook KLM because it gives you a lot of different things that helps you to look at data differently.

So you can see here, this is producing requirements this is then using that to then generate more content and it's absolutely phenomenal right if you think logically and think through it it can really really help you get through stuff so quickly and build your requirements really quickly so let me just go back to the um so these are a whole bunch of the stuff a whole bunch of the things so as an example

Applying AI Across the Business

We are, you know, we're doing some of the marketing stuff at the moment to build out an AI first consulting value proposition. And this, you know, what we've done here is we've taken all of the different documents that we've got from all sources across the web and some internal documents and a bunch of other things and are asking it to produce the website, the content, the value prop, the marketing plan, et cetera.

So we do this all the time. We do this for everything, right?

It is phenomenal what these tools can do. And this is just an example.

Engineering Complex Systems with AI

So here, this is, I mentioned earlier about we're building using, building a system that allows to build software. So you can see here, I did this ages ago. This is about 50 days ago.

And this is a paper that we're using to help us understand what are all the aspects actually building sophisticated software. and moving away from just doing vibe coding, you know, lovable or whatever else. If you want to build systems end to end, it's a lot more complex to build systems that companies are going to use, right?

And so being an architect, I understand in my head what those things look like and I take that out and how do you then transition to reinvent how you do requirements.

So you could see this is an element requirements engineering. You can see just how it does it and how you can explain it.

Communicating with Customers

This then also helps with customers. So when you talk to customers, you can use a PowerPoint deck with that, and you can go and help customers figure out how they should do it.

And this applies to anything you do. If you think about the way that, going back to the ways of working, think about how you can use it in every step in the process.

Best Practices: Don’t Rely Blindly—Iterate

But don't rely on it 100%. Don't just think, well, let me do it. And don't be lazy and just say, oh, let me produce the output and thank you very much.

You know, iterate through it. Build a structure. Push that through.

Prompt Libraries and Reuse

Learn prompting. Now, prompting sounds easy and you'll see lots of prompt structures out there. And there are many, many prompt structures in many different areas. Find the ones that work for you and store them somewhere and use those to help you produce stuff faster.

So for me,

Becoming AI‑First Individuals and Teams

What we're seeing and what we're doing internally is everybody we're hiring We're basically putting them through this approach to be an AI-first company and therefore an AI-first individual in everything that they do, in everything they output, in everything the way that they think, because it helps us innovate faster, helps us deliver faster, work with customers faster.

And for you guys, if you do that, frankly, you'll be in a way, way better place in terms of the kind of job that you'll be doing and doing now, but also being, frankly, more

Career Impact and Hiring Approach

um what's it more skillful in what you do so there we go so i'm happy to ask answer questions so uh so i'll take the one that says thingy engineering first and it says especially now it's fascinating how do you prioritize a source to focus

Live Q&A Highlights

on the new still trial and tested approaches is that down to manual creation um no so yes you can do manual creation because you can say you know what i know these 10 documents roughly i can import those but use llms to do it you can use llms you can use gemini a bunch of other tools say yeah i want to do this which of the papers should i use to um

do what i need to do so it's a month combination of both it's not either or it's about finding the best techniques yourselves and get the large language models to help you do it don't try and don't think i don't know how to do this ask the question and it will help you learn that and iterate through it and you'll find the best way to do this particular task so so this is not either raw it's finding the best set of tools and the processes

um but changing the way you think work and say well how can a model help me do this what do i need to tell the model and how can i get the best out of the of the llms and the tools like um notebook lm which is what i'm showing you here to get the best result um

Indexing, Search, and Knowledge Graphs

Chris is asking, how do you practically query the huge knowledge base? Do you index and research documents? So yes and no.

So in this case, Google Drive, it does indexing for you. So you can find those documents that way.

But part of the platform that we're building is to do the indexing documents and be able to build what are called knowledge graphs. I didn't talk about knowledge graphs, but you can build structures that allow you to connect all the documents together

then you can use that to to be able to find the documents you need so there's different mechanisms and tools that you can use to actually do not just a standard search because standard search is okay but using llns but also building structures um i was going to show you something called the rag process if you know what rag is any of you regular augmented generation which is what you do

RAG Basics and Vector Databases

take your knowledge base you convert that into a specific format and there's a bunch of things in the middle which i'll bore you with in terms of how to do that and you have what's called a vector database and then you can attach that content to a model and use that to then have your specific bigger context to be able to um use that as part of the uh as part of what you do um

Cataloging: Tags vs Knowledge Graphs

what is the best way to cataloging content so easy to use in both internal system mapping like mcp and ai driven tools like rag do you follow a specific framework approach so there are many different approaches to that so mustafa's asking um so cataloging content so one way build a knowledge graph and that will enrich all the content and frankly you can do it dynamically right you can take your content you can tag it but tag it through a knowledge graph and that gives you all of that content already enriched you can build a metadata database you can ask an llm to do it for you because it will do it right so you can take all the content and say here's all my categories and tags tag all my documents etc and i don't like what i call linear tags which is a tag which is just the word doesn't it doesn't really give you the depth you build a knowledge graph and again i won't bore you how to do that it's there's tools out there to do it but if you build that that's way more effective than just building linear tagging so most things right now use metadata tags and tagging i think doing um uh building a knowledge graph is way better um

I was going to ask a thingy. I was going to ask a related question to Chris.

Managing Knowledge Graphs in Practice

If you're using semantic search via vector DB, yes. Yeah. So this is part of the work you need to do, right?

Knowledge graphs are actually quite complicated to build. Some of the tools that we've built allows you to dynamically build knowledge graphs.

And part of the work we're doing is to build a system where you don't have to understand how knowledge graphs are created. You just use the tools that we're building and it will create it for you. And so you'll have that base.

So it's technical. How do you manage? Well, yeah. So Nicholas is asking, how do you manage the knowledge graph part?

You need a bunch of tools. That's why we're building what we're building, because to actually use that, create it, manage it, update it is a reasonable amount of work.

So.

Applying Patterns to Real Problems

Next question, Brett is saying, so if someone wrote a paper in 2005 on how to find exoplanets and machine learning and wrote some code and then gave it up and gave up and out, could you help us find the best way forward?

Yes, absolutely. Right.

Yeah, I mean, this is the kind of patterns you need to think about. I mean, this is what I said earlier about ways of working.

If you understand the steps you can do, you can solve so many problems. You can solve and do so much stuff that you really couldn't do before. You just have to understand how to do it.

Tools of Choice and Trade‑offs

So Nicholas asking, Geo4j, yes, that's one graph database that we've used. Yeah, I mean, there are others out there. Um, there are, you know, pros and cons with Neo4j, but at the end of the day, yes, that's one of the, one of the, one of the databases we use.

So, so any more questions, um, you're asking me anything, um, I'm here to try and help you guys become AI first individuals and work with AI every day and everything you do.

Why Build Your Own Database vs Relying on LLMs

So Andy Hickman's asking, what's the advantage of creating your own database of docs by just relying on what LLMs were trained on? Is it because many of those papers are not in public domain?

No, I think, Andy, for all of you guys, it's more about having an extension of your own brain, of your own content, right? If you're writing a document and just putting it into perplexity and say, write it for me, where's the USP?

So building your own database of your own content, I tell you, makes a massive difference because you're creating and curating unique output, which you wouldn't get if you just went on to perplexity and did it. Or some other tool or a chat GPT or some other thing.

Start Small and Iterate

Do you start big and narrow down or start small and spiral out? You start small. You iteratively build it.

You iterate, iterate, iterate. Can't say the word. I can't say the word.

You know what I'm saying? Iteratively build this stuff yourself. You build your knowledge base.

Build your content. Build content and use that content to build more content, right? So, you know, you can build your own stuff.

Leverage Open Research to Differentiate

So the kind of things you can see that we've got here, we then use as part of our base content and feed it back in, right? So it's absolutely best to do it.

And you can find research papers on anything out there, all open source, all available, and bring them into your knowledge base. So I showed you some of our knowledge base, which is Gen AI research papers. We build that context because that gives us a USP in terms of how we can then use that to build our requirements.

Bots and Agents in the Enterprise

Do you use bots integrated into Teams, Slack at all, easy to make information?

Yes.

Turning Knowledge into IP and Methodologies

I mean, the key thing here is if you're any kind of size of business, you're a small or large business, and I think the chaps before talked about it, one of the key usps of the business and international property you can build is all the knowledge you have across all parts of your business if you don't have it bring it in bring it in that will give you the usps of your business um and frankly you can so if you do consulting one of the things consultants do a lot right is they come in with their methodologies hey we've got a unique methodology of doing something and they'll stick a bunch of powerpoints on the screen and go this is how we do everything right i tell you what Until LLMs came along, that was IP.

Today, you can build your own methodology and your own process. Right. And you can actually even better than that.

You can say to the customer, hey, tell us your business in your context. We'll customize it for you and we'll produce a whole methodology and we'll give it to you. Right.

So people like McKinsey and Bain and all these other consultants walking with their methodologies, those things are all available either to build yourselves and use as part of the work you do for customers or use to improve your internal or your own internal working and working practices.

No, I can't show any bots at the moment, but but.

Agents Across Business Functions

But the idea is, you know, one of the things that's going to happen, right, is you will have, wherever you work, either you build, if you're in a small or medium-sized business or a lot, you know, obviously bigger businesses, there'll be IT and everything else, is there will be agents running to do almost everything across every part of the business.

So if you're a small business and you're building, I don't know, let's say you do marketing and agency work, you can really quickly use agents to do everything that you do. You can do it. You can do it yourselves.

There are some tools and companies out there. And I would think about those. But the important thing is that this stuff is available and you can.

One of the key things to help your business grow and help you make you more efficient, et cetera, is really, really deeply learn this stuff.

Mixture of Experts and Advanced Techniques

Andy, also, LLM weights will be generalized, so always worth having your own RAG content to give more specific. You can prompt to tell an LLM day as a spot expert, which you can help focus on. Yeah, mixture of experts.

Absolutely. Yeah, absolutely.

So mixture of experts, if you don't know, I mean, I can... The thing is here, there are so many techniques that are out there now in terms of creating content, producing research, doing analysis.

Generating Analytical Models and Spreadsheets

There are LLMs that actually you can use to actually do complex mathematical calculations, right? Or build your spreadsheets or build your models, build the spreadsheets models, right? And I've done that.

Some of the stuff that I'm working on, I literally asked it to generate whole model in the spreadsheet for me. And it did it. It did it.

It did it. You know, I didn't have to spend weeks, days, weeks trying to do it. I just said, build me this.

This is what I want. This is how it's going to work. And it did it all for me. And it did it in about 10 minutes.

Prompt Craft and Rapid Iteration

So, you know, the key thing is knowing how to create the prompts, understand the mechanics of how you could do it and give it a go, right? There's no harm. Just give it a go and iterate.

Keep iterating through it to get the right answer.

Cool.

Wrap‑Up and Final Tips

I think I'll stop because I've spoken for about 20 minutes. I'm hoping that I've inspired you guys to think about using AI for everything that you do and everything that you learn. So one of the things that I did here was just to show you.

Where is it? Where is it? Where is it?

Let me go back to Notion and then we go back to here. So one of the things that...

Build Your Own Training Courses

One of the things you'd also do is build your own training course to learn how to do things. So this, this is, I literally did this while, while the other chaps are talking, I did, okay, you know what, let me build a training course for myself to do how to learn how to use Genio. And I literally, I did this and you know, one of the key things, because it's so amazing what you could do, you could build your own training course on learning how you want to do it in the way that you want to learn.

Yeah.

So you don't have to go out and go to these generic training courses. You can build your own.

Reducing Hallucinations: RAG and Citations

Sorry, the question there was, do you have hallucinations when you use dedicated knowledge base? No, you don't get.

So one of the key things is if you're doing the RAG process, behind the scenes in Notion, you don't see the RAG stuff, but they're doing RAG. The challenge, however, with RAG is they can only do text. If you want to do video content, you want to do some other kind of content, it's not necessarily great because you have no control over what Notion is doing at the back end around RAG.

That's the point around building the RAG process, a part of the tooling we're building. is to enable you to control the different types of rag process to build the different types of um uh content so for example mathematical mathematical formulas from a paper generalized model is not really great at that you can use build a rag process with a bunch of things and use a model to then build complex formulas and complex mathematics that that's something that you could also do i've created documents about me like writing things i'm concerned about it skimming and missing things

um so one of the key things that you have to do is use citations so in perplexity it's really great doing citations if you tell the llm give me the citations it will do it yes that's one thing to do secondly is it missing things and skimming things this is where the human in the loop comes in Right. This is where the human loop comes in.

You should read it. You should understand it, given what you're doing and iterate through it. So if you do that, then you won't you won't get you won't it won't skim things or miss things.

Right. If you if you know what you're up to.

Human‑in‑the‑Loop and Data Science Techniques

That gives you way better. That improves the iteration, iterating through to get better answers with the output from LLMs as well.

Choosing the Right RAG Variants

to get the better results that you would need uh um and i just i've tried not to go into technical depth just because there's a lot more like i'm some of the some of the audience have mentioned rag um rag there's about 50 plus different rags rag approaches that you can use understanding each of those ones and which ones you apply for which use case, these things are complex. So most people wouldn't really know how to do that.

And that's why you build a platform when you do it, because you then hide away that, you know, the user says to you, I want to do this. And you say, okay, here's the right processes, which one you could use this one, this one is the best one.

So for me, this is the kind of stuff that we're trying to simplify, because to get the most effective output answers in the productivity, You need all this stuff inside, but it's hidden away. The complex is hidden away for you.

Hiding Complexity Behind Platforms

Cool.

I think I'm done. Unless there's any more questions, I'm done.

Conclusion and Contact

Please reach out to me if you want. I'm on LinkedIn. Just let me know.

If you connect to me on LinkedIn, just let me know that I did this webinar, etc. So you know that... Uh, so I know that that's where you connected.