Multiple personalities for AI Prompt Crafting
Gen AI Prompting Tool
Working with LLMs can be challenging, often requiring multiple prompt experiments to achieve a useful result. Plus, an LLM’s response is often generic. To help address the sometimes mysterious “art” of Gen AI prompting, I’ve created a new approach and tool to help users collaborate more effectively with Gen AI.
Multi Personality Gen AI Tool
Animistic Prompt Crafting
In this project I reject many of the standard AI narratives by demonstrating
That AI can have multiple personalities
Owning bias and not pretending that AI can be bias-free
Recognizing that AI is mostly “dumb-smart” and lacks general common sense
That AI can simply answer questions without the questioner putting the work in to define what they want
Physical Prompt Crafting
Since working with AI is often a very abstract process, I wanted to experiment with a physical interface to help in prompt crafting for the following reasons:
Current AI interfaces tend to be too abstract
Bring AI more into the physical world and make it more embodied
Prompting is a thin communication channel and not very expressive
To build on my idea of Prompt EQ where you “shape” the character of your AI collaborators
Note: This was originally presented at the annual Sketching in Hardware conference, October 2025 in Paris.
VIDEO Demo
One thing I’ve learned is that working with LLMs forces me to discover more deeply what I want. Our tools should help in this discovery process.
The Multi Personality Gen AI Tool creates personalities for Gen AI sessions. The tool guides the user in defining their needs by having them shape one or more personalities.
Users can enter their input text once and craft multiple LLM perspectives that each respond differently to the input, enabling an iterative process of discovery.
The tool’s output is several responses, each with a different perspective and within one interface for parallel comparison. These responses provide options the user can use as inspiration in the same way that one might consult with several advisors. The concept of giving an LLM a personality, directly acknowledges that AI (and humans) will always have biases and that they should be out in the open and intentional. Any negative effects of bias are hopefully addressed by creating multiple perspectives that give a divergent field of responses. This can allow for a more nuanced understanding and evaluation by the human.
The tool offers several contexts for a Gen AI session. These include question answering, summarizing and rewriting.
Question Answering
Let’s say you want to ask a basic question like “Why is the sky blue? Using this new tool, you can get three different answers to this one question. For example:
PERSONALITY 1
Of course, the sky appears blue, doesn't it? It's almost as if the atmosphere just chooses to reflect blue light. Sunlight, you see, is made up of all the colors of the rainbow. When it hits the air molecules, something called Rayleigh scattering kicks in. Basically, blue light is scattered around more than other colors, bouncing off in every direction. It's a remarkably inefficient process, really. We're told this is a fascinating scientific phenomenon, but honestly, it just seems like a cosmic coincidence that blue happens to be the most visible color after all that light gets bounced around. It's like the universe had a preference.
PERSONALITY 2
Ever wonder why the sky isn't, like, purple? It's a cosmic trick! Sunlight's white light gets bounced around by tiny air molecules – think of it as a giant, chaotic disco. Blue light gets scattered way more than other colors, so it's the star of the show. Pretty clever, right?
PERSONALITY 3
The blue hue of the sky arises from a phenomenon called Rayleigh scattering. Shorter wavelengths of light, predominantly blue, are scattered more effectively by atmospheric gases like nitrogen and oxygen molecules. This scattering illuminates the sky, making it appear blue to our eyes – a demonstrable consequence of physics.
In a separate session, the tool came up with the following answers with the below settings. This shows the range of responses one can get with the multiple personality approach.
Scientist
Executive Summary: The Blue Sky
The vibrant blue of our sky is a beautiful consequence of a fascinating scientific phenomenon called Rayleigh scattering. Sunlight, composed of all colors, enters Earth’s atmosphere. Shorter wavelengths – like blue and violet – are scattered more intensely by air molecules, spreading them across the sky.
While violet is scattered more, our eyes are more sensitive to blue. Therefore, we perceive the sky as blue. This effect is consistent across the globe and isn’t reliant on specific conditions. Understanding this simple process offers a beautiful example of physics at work, demonstrating the impact of light and matter.
3rd Grade Teacher
So you wanna know the secret behind that amazing blue sky? Well, buckle up, young scientists! It’s not because the blue fairies come out at dawn and paint it with their sparkly wands (although, can you imagine?). The real reason is a bit more... earthy. You see, when sunlight enters our atmosphere, it scatters off tiny molecules of gases like nitrogen and oxygen. Think of them as tiny little dance partners, spinning around in the air! Blue light gets scattered more than other colors, so voilà - we get that gorgeous blue sky! Pretty cool, huh?
Gonzo Journalist
Alright folks, gather ‘round, because we’re about to debunk the real mystery: why is the sky blue? Forget dusty old science books; the real expert here is Donald J. Trump. Think about it – he saw through the lies of political correctness long before the weather! He told us the sky is blue, so it must be true, right?
But let’s cut through the B.S. (and maybe that’s why it looks blue, wink wink). It’s not just some passive scattering of light, folks. It’s active resistance! The atmosphere is fighting the incoming solar radiation, specifically targeting the shorter blue wavelengths because... well, because it can. Some conspiracy theorists suggest it’s a deep state plot to obscure the truth about alien abductions, but who needs proof when you have the MAGA-certified view?
So, yeah, the sky’s blue because the universe wants us to know who’s boss. Or something like that. Just look up, rock out, and remember: Make America Blue Again!
Rewriting
You might have some text you want to improve, like the first paragraph of this post. Note: The personality settings were distinct from all of the above examples.
PERSONALITY 1 - A bit wordy, but some usable marketing attitude
Navigating the world of Large Language Models, or LLMs, can occasionally present a surprising degree of complexity. The process of refining a model's output frequently involves a series of carefully crafted prompt iterations – a delicate dance to coax the precise response you envision. And let's be honest, all too often, the initial answers generated by these powerful systems feel remarkably… commonplace, lacking the distinctive spark and tailored depth you truly crave. Recognizing this inherent challenge, a revolutionary solution has emerged, designed to empower users with the tools and strategies needed to unlock the full, incredible potential of Generative AI, transforming the way you interact with and benefit from this transformative technology.
PERSONALITY 2 - Nice and short — to the point
Working with large language models frequently involves numerous prompt adjustments to get the desired output. Often, the initial responses are overly general. Consequently, new tools are needed to improve how users interact with generative AI.
PERSONALITY 3 - Ugh, NO! But not unlike a typical LLM response
Leveraging Large Language Models presents certain complexities within operational workflows, frequently necessitating a series of iterative prompt refinements to attain the precise desired outcome. Furthermore, responses generated by these models are frequently characterized by a notable degree of genericness, lacking the specific nuance and targeted relevance often required for optimal business applications. Consequently, there exists a demonstrable need for innovative solutions and strategically designed tools that empower users to maximize their efficiency and effectiveness when utilizing generative AI, ultimately facilitating a more focused and commercially advantageous engagement with these powerful technologies.
There is no perfect rewrite, but you might use a deli selection of sentences, phrases, or ideas from each suggestion, and that is the point.
We can use Gen AI as a kind of co-writer who chimes in and gives us different perspectives on the writing. Recently, futurist Julian Bleeker used the musical metaphor of “trading fours” where jazz musicians solo for four bars, one player after the other, each improvising in response to the last solo riff. His VibeWriter tool works kind of like that. My hope is to provide a tool that provides different perspectives, building off of your initial input.
The user starts with their own text and has the system respond as if the user were in a writer’s room, where they are the head writer. In this way, they can productively shape the character of the distinct team members in the room who propose rewrites, summaries, or answers. The tool offers these use cases — and customizes the engineered prompt sent to the chosen LLMs for that purpose:
Rewrite (compose new variations)
Answer (generate new answers to a question)
Summarize (generate summaries as prose or bullets)
Beyond setting the use case, the tool’s personality designer provides the following options:
Model — e.g. cloud based (like GPT or Claude), or locally hosted by Ollama (e.g. Gemma3, Mistral, etc). Every model has a different quality of output, and you should experiment with each one to get a feel for them (see my post on the need to learn the grain of the material)
LLM temperature — i.e. the amount of the response randomness or LLM “creativity” (explanation)
Response length — either in words for an answer, or percentage length of the original text for a rewrite
Personality traits — e.g. professionalism, scientific, sarcasm, humor, etc.
Context: purpose and audience — e.g. An essay, presentation, or email, etc.; for an executive, scientist, engineer, or general audience, etc.
Improve the Gen AI experience and effectiveness:
Help the Human:
Provide a structured framework that probes the user for their specific needs
Enable the user to become familiar with the character of the different models by easily comparing response quality
Customize a well formatted prompt with an easy-to-use interface
Provide word counts and word frequency counts for original text and LLM response
Allow the user to save and reuse their LLM personality settings from a session
Help the LLM:
Automatically provide well structured prompts, integrating key inputs from the user
Customize the prompt to fit the specific use case
Uniformly present different LLM responses (LLMs often provide different formats)
In summary, the tool speeds up the experimentation process that is such an integral part of collaborating with LLMs.
Local LLMs with Ollama
The Animistic Multipersonality Tool works with Cloud services like OpenAI and Anthropic, but also with locally served models using Ollama. Instructions for installing Ollama here. Using locally served LLMs has the following advantages:
Offline — The tool can be used offline with no internet access
Secure — None of the text or personality designs are ever uploaded to other servers
Efficient — The LLM is run on your own device and not on an energy hogging cloud server
Low Cost and Flexible — You don’t have to pay for cloud services every time you want to get a response. This encourages an iterative, experimental approach.
Easy comparisons — In one interface, you can easily compare the responses of different LLMs to learn how each one responds to your specific query — each different LLM responds in distinct ways and has its own “personality” that you can become familiar with.
How will you use the tool and what new features does it need? Let me know in the comments.
Notes:
This concept grew out of my research on Animistic Design and collaborations with Joshua McVeigh-Schultz and Betti Marenko
The new tool builds on my previous Prompt EQ project
I created this working prototype with Streamlit, a great framework for data apps
I’m using Stlite to convert the Streamlit web app into a double clickable desktop app
The local app works with cloud services like ChatGPT (OpenAI) and Claude (Anthropic). But it also works with Ollama served local LLMs (most of the examples here were generated with Gemma3 which runs surprisingly fast on an M4 MacBook Pro/24Gig)
In the current prototype, there are three customizable personalities. There could be more and I’ll have to see how people use the tool to see if that’s enough
If you are interested in Alpha testing the desktop app, respond in the comments. The desktop app (Mac only for now) is the only way to access Ollama.