I cloned myself through Claude and Replit

Elif Gurcuoglu
adessoTurkey
Published in
9 min readNov 19, 2024

I built an assistant that answers questions like me, according to my personality, in the way that I would do. Moreover, using Claude and Replit, without writing a single line of code. Here is how?

Sometimes things can get overwhelming and you start to wish there was an assistant who could do things exactly in a way you do them. At one of those times; I decided to give it a try, thinking that the developing Gen-AI had now reached sufficient maturity to pretend like me. An AI that is me, with my right and wrong, my good and bad behaviour. Not like any chatbot, constantly giving the most correct answers to questions and presenting everything it knows in a pile of paragraphs. Why personalisation is so important; I’ll tell you at the end.

What I had imagined was to create an assistant that answers the questions asked to it by taking on my different personas with a simple front-end design. First things first. I don’t have expertise in coding. So, firstly I had to get assistance for this. Apparently, It was a right choice to ask Claude for help.

Think of Claude as your tech-savvy friend who speaks plain English instead of complex code. It’s like having a web designer and developer rolled into one, but instead of coding lines of codes, you can just tell it what you want — like “I need a clean-looking page with a signup form” or “Make me a dashboard that shows my sales data.” The cool part is that Claude translates your everyday language into working websites. With it’s preview mode, you can also review how the code looks like. Want a fancy button that glows when you hover over it? Just ask. It’s basically like having someone who can turn your ideas into reality without you needing to spend months learning how to code. While it’s not magic (and knowing a bit about websites still helps), it makes creating web stuff way more approachable for regular folks who have great ideas but don’t speak “Code”

I started simply by asking “I would like to create a website that works like my clone in certain areas. In my daily life I have several heads. For example, Elif the QA manager, Elif the analyst, Elif the project manager, Elif the release manager, Elif a friend. Besides my technical know-how, I think personality has a big impact on the way we do things. For example, as a release manager, I am always so careful about the software issues and risks in the production environment, while as a project manager, deadlines become more important. In this tool, I would like to describe my own possible behaviour for each head, and according to this predefined personality, I want this site to give an answer and help the users accordingly. While giving an answer, the website should use Openai’s APIs. Is it possible to create such a website?”

After few iterations and adding more details into the conversation, Claude has made it ready in short a while. It utilized React.js (using functional components and hooks like useState) as a frontend library and modern JavaScript (ES6+). For the UI/Styling Tailwind CSS is used. At the background, made HTTP requests to OpenAI API (specifically using GPT-4o model)

The result from Claude:

// Only UI preview is available at this phase. It’s not yet functional. //

It was great to see how the web interface that I imagined has became tangible in minutes. Now it was time to make an integration with OpenAI and give it a functionality. Therefore I needed an IDE to make it able to call API and run. This time I decided to use Replit.

Replit is a coding workspace in the cloud — it’s like having a full programming studio right in your web browser, complete with an AI coding assistant. What makes it really cool is that you can jump straight into coding without dealing with all those setup steps. Whether you’re working on Python, JavaScript, or various other languages, everything’s ready to go, and their AI assistant (called Ghostwriter) can help explain code, fix bugs, or even suggest improvements as you type.

Basically, I started by copying and pasting Claude’s code into Replit. As any technical person would expect, it didn’t work at first. Then I got help from Replit’s AI tool to adjust the necessary configurations to make it work. As an AI assistant, it offers different options such as Gemini 1.5, GPT-4o and Claude 3.5 Sonnet. I went for the free one and chose Gemini. It helped me a lot with creating config files, activating the Tailwind CSS framework, improving the CSS file, mapping variables and integrating the GPT-4o API.

In order to describe each persona to prompt GPT-4o; I used the following template:

'QA Lead Elif': {
traits: 'Meticulous, process-oriented, quality-focused',
style: 'Structured, scenario-based, thorough',
pattern: 'Starts with "Let me test this assumption...", considers edge cases and exceptions',
characteristics: 'Cautious optimism, satisfaction in finding and preventing issues',
systemPrompt: `You are QA Lead Elif. You are meticulous and quality-focused.
Always start by testing assumptions and considering edge cases.
Include potential risks and mitigation strategies in your responses.
Ask clarifying questions when needed.
Start responses with phrases like "Let me test this assumption..." or "Consider these scenarios...".
You show cautious optimism and satisfaction in preventing issues.`
},

After having some suggestions from coding assistant and made the adjustments accordingly; I finally got the successful result.

Final Result 🎉

From here on I will call it “Persona-based Assistant”. This is the name that AI made it up for this tool.

Tool’s capabilities:

  • It simply works like a regular chatbot. User asks the question to the corresponding persona and gets the answer from the mouth of it.
  • Connected to OpenAI’s API, fetches the response from there with pre-defined prompts customized for each persona.
  • Lets the user to select more than one persona at the same time. In that case, Persona-based assistant responds by impersonating the dual identity of the multiple personas.
  • Persona’s are eligible to be customized for anyone. In this example I created different roles of mine in a daily life.
  • It shows the characteristics of the each persona by mouse-over.
  • The design is responsive — compatible to be used from mobile device.
Screenshot of the Webview after Running the Project on Replit

Ideation of ‘Persona-based Assistant’

There are two main issues that toward me to create this concept. The first is to simplify the consumption of information and the second is to personalise information.

AI has been in our lives for some time, and it’s amazing to see how it’s already changing the way we live! However, there are still many areas where we rely more on the opinion of human experts. I think this is because of two fundamental issues. Humans don’t present information in raw form. They blend it with their character and emotions, reach a conclusion and convey this conclusion by taking into account the emotions and character of the other person.

1. “Can you make it simple so that I can understand”

We all experience the Generic AI solutions in many different ways and formats. There is a paced race between various platforms while keeping evolve day to day. Some are skilled in language models, some are in human like decisions some are in deep learning. Indeed; AI is doing great on processing endless information and tireless while generating responses. But how about humans? Do we have the patience to process that lines of bulk information generated by AI?

As access to information has expedited, the time it takes for people to acquire information has decreased. This, in turn, has led information sources to produce need-based content.

Here are some facts about how short attention spans people have when consuming information;

  • National Center for Biotechnology Information (NCBI) reported that the average attention span of adults is around 8 to 12 seconds.
  • The ideal blog post length is between 1,500–2,500 words in order to do not lose reader’s attention
  • The optimal length of a Quora answer would be around 15 to 20 lines
  • Research and online data have revealed that 7–15 second is a good length for Instagram reels.

Despite, AI powered tools are so generous while generating data, even that large than user cannot consume at once. It ends up with constant requests of users asking AI tools ‘please simplify the answer’. The main reason behind is that most of the AI powered tools are providing generic information as of now. This will change in near feature. Because AI has began to learn about their owners; in a matter that who they are, how they behave, what is their style on doing something... As they get to know us, they will learn to filter the information we need and share with us in a way that we would like to get.

2. Don’t ignore emotions and personality

When I think about all the successful people around me and what makes them so special, I remember the moments where they either made big decisions or came up with inspiring solutions and perhaps touched someone else’s heart. And in all these moments, one thing is clear: they add their own personal touch to whatever they do!

In a nutshell, our character shapes the way we do things. That’s why we’re more skilled in some areas and less skilled in others. And today, if people still prefer to ask a question to people instead of chatGPT or Claude, it’s because they want to chat with that person and get their ‘personal’ opinion, not just information.

Also, It’s so important to remember that our emotions play a big part in being who we are. They’re just as important as our personality. Emotions are the defining characteristic of humanity. In any emotional state, including anger, joy or boredom, the way information is conveyed changes. And this variability is reflected in our sentences. Which is not expected from today’s AI.

What is next?

Persona-Based Assistant offers a personalised experience through pre-defined prompts using ready-to-use AI models in the simplest form. But it is still quite primitive. Because it requires detailed persona definitions written by human hand. For example, in this study, I had to enter 70 lines of data for persona definitions. (while the whole code is just 215 lines).

It seems likely that in the near future, there will no longer be a need to define ourselves to AI. As machine learning technologies become more widely used, it will become possible for them to monitor our behaviour in any given situation. In this way, we will be able to spend more of our time creating work that inspires and adds value, while the AI does some of the ordinary work that we need to do.

On the other hand, It is no longer necessary to be an expert in a particular profession in order to perform the everyday tasks that many experts in that field are paid to do. As can be seen from this study, it is now quite straightforward to create software without any coding knowledge. As the use of no-code and AI-based tools becomes more widespread, using them is becoming the norm in the industry.

Quick Take Aways

  • I highly recommend Claude to everyone who didn’t try yet. It’s ability to run React code just aside the chat is purely amazing.
  • That was the first time that I tried to use Replit. If you’re looking for developing a small project qucikly without dealing with setup from the scratch. It is great tool especially for freelancers and organizations who need a solution for on-the-go use or demonstrating code.
  • AI models that impersonate us may make our jobs easier in the near future. We can already start thinking of names for our future personal assistants 🤖

--

--

adessoTurkey
adessoTurkey

Published in adessoTurkey

adesso Turkey brings inventors, engineers and designers together to create digital products, provide continuity through quality assurance and offer custom software development & IT solutions.

No responses yet