Introduction

Yeah, we will overcome bureaucracy today.

I will show you four solutions.

Meet the Presenter

Here I am. I'm Peter Senkus. I have quite some roles. You can read it.

The most role what I like it is to be hacker of bureaucracy and to simplify things.

Overview of the Presentation

I also asked to connect my friend. He will connect us and show us ready working solution. My student will come to us from Katowice and we will present you how to

Aims of the Project

using AI for project briefing and claim system so I will show you how to make brief of the project because I work also as a solution architect and we will show you ready solution or four ready solution to overcome bureaucracy so what we have we have some tensions here

Rapid App Development with AI

It is current state, we have some work to do and actually we will be happy in a couple of days, couple of weeks. More couple of days because developing app nowadays with AI it is quite one day or two days.

1Imagine to be solution architect, imagine draw that stupid UML diagrams how many days you did it. I am 20 years in it and I know how to do it.

The Problem with Bureaucracy

When you write to Vice Dean of Student Affairs to one of the Polish universities you will get autorespondent like that. Imagine 30 days. Polish rules are that you should get it in 14 days. And imagine you have a case that you want to act immediately.

So I will share with you also all of the brief what I prepared. I will show you that. And I will show you my prompt library to prepare it.

The AI-Enhanced Project Brief Template

I did some Excel everybody is looking for Excel in company so we have Excel with project brief you have purpose definition you have initiatives you have goals that is checkpoint basically when you have problem then you have who is affected by problem, then how does the project goes into organization goals sometimes we need it especially to incorporate how to go to strategy what are the risks here we have stakeholder analysis because not only shareholders then we have activities brief then we have user personas then we have user stories so basically it goes all agile then Moscow we need it when we are designing an app then we go to yeah that is Moscow then we go to Perth

So let's see how long it will take us. Basically, I asked ChatGPT with normal way. It told that the classic way to develop such app will be over 100 days. And that app was developed in three days.

Augmented Intelligence and Development Speed

So what is power of AI? I don't call it artificial intelligence. I call it augmented intelligence.

so here we have RACI matrix if we work more than one person actually I was using that in my lectures so we had four people group so they did RACI for themselves too they developed not that app but similar app to leading Polish companies I have NDA so I cannot name them I can name one Neutro Pharma and for different marketing purposes and so on and so on.

Agile Documentation Essentials

Then we have service proposal and Gantt and then we finish with risk management. So basically that's all what is needed by agile documentation.

Then you have diagrams That's cool because you can do it really smart way you generate with ChatGPT then you go to some page which generates it, I will show you The thing what I am using for it is DrawIO

Live Demonstration of AI Tools

One moment I will switch to another vision so I can demonstrate you how it looks live. Now you see my screen.

Basically I will replicate that UML diagram, just I need only the description of the app. And here is my prompt. case diagram in a plant UML base on description. So we provide the description and it will provide us nice plant UML hopefully

yes it is yes I think we need is to copy and we go to draw.io it's nice free tool I like open source and we go to draw diagram, let's say in brackets.

Yes, we have nice UML diagram with help of AI. So to design such app It took me one and a half hour to have first draft. Of course you can make it better and better, no problem. But yeah, it is the first draft, it is like charm.

Deep Dive into Solutions

So let's get further and I will go with you through the solutions. Of course, I will show you with prompt library. I have also all the prompt library you can go through how it was done.

couple pages also with some errors because chat GPT or cloud is not 100% ok, sometimes you have to push to get desired outcome.

Implementation Phase

So let's go to the implementation phrase. do it only with pre-trained model it is not good solution because you can automate for example with make that it will customer claims and students claims and answer with knowledge of chat GPT it's horrible idea because it concatenates it has a problem with data sources and so and so on

1So what we do is we combine with RAG, retrieval augmented generation, We minimize hallucinations, we make content adoption and we reduce training needs and it's quite scalable solution.

Data Preparation and AI Responses

If somebody likes diagrams, it looks like so, but first we are preparing data source, then user takes a query, and receives response based on the data which are here.

I don't agree that AI is a black box, it is not. OpenAI is a black box but when you download LAMA with Mixtrall you will have like 55 line of codes and 100 GB of data so it is compressed per internet so what we need to use LLM only to script the data first solution what you can do you can make your custom GPT with OpenAI

Custom GPT-3 Model Creation

actually that is ready yes here we have that is crafted with the knowledge here is a construction site I have special formula to craft the model. I write mission, methodology, what it can use and some assumptions

I did it like a lot of such models, when you put the data jacking up, here is basically study rules, some financial help and so on and so on or what is needed for study students

Yes, if you want to check if it works, no problem. You can scan and ask for some questions. You can check in English, Polish, Swedish, whatever language you want. I think I did a rule that it translates for English.

Automating Email Responses

Ok, but that we have local, that we can put on some web page, even through some user interface, but how to make it to autorespond email. Also you can make with make actually from Integromat.

So basically also in that presentation I will not go through all. You will have step by step how to do it. Here you have all of the things how to do it, how to go through it.

So email comes when email is assigned in the Google sheet. But at the end I will show you only process, how it looks. That is a process.

End-to-End Process Overview

So we have incoming mail from the students with some claim. The claim is registered in the Google sheet. Then it goes with HTTP to your private GPT. then it answers the question and it emails back the question to the student or everybody is interested in it.

So that was the solution. Second, I don't like it either because it is on the web. I like on-prem LLMs and here we have in general risk with LLMs.

Data Security and Local Solutions

About 90% of people using ChatGPT

are sharing very flexible data security data from the companies it is not quite good so we have another solution which can run locally and Timoteusz is with us just show because we cannot hear you so

Automation with Make and Flowwise

So that is the automation in Make, like I told you. It assigns all of the documents, incoming mail to the web server, to the Google Sheets. Then it goes to the Flowwise.

Then the response is also crafted with ChatGPT, written down to the server. Then it goes back to the student or everybody else.

Next.

Demonstrating the Automated System

that is FlowEye site it is also drag and drop all the process in it and then Tymek please show the outcoming mail because we are running out of time so show the product how the automation works now it's running and then we will see the mail which gets to the student yeah it's successful that is response from the chat GPT

It gives all of the data and then that is the outcoming mail. Of course you can format it but it requires a little more time.

Thank you very much. Also we can provide a link to it and how it's done.

Building a Powerful On-Premises Server

What I'm really into it is solution 4. I'm building own server. It has 3 graphics cards RTX 3090 so it's quite powerful to be in small business. If you want more information about, please ask.

Conclusion

I'm 20 minutes, so five minutes after time, but I think it was quite interesting.