From Simulated patients to MECHABETH
Why I built a chatbot version of myself and how you can too
Throughout my career, I have embarked on many different digital projects, but perhaps the most personal digital project I’ve worked on is MechaBeth, my own personal chatbot trained to simulate my own responses. It can confidently relay my tone of voice, my experiences and my approach to teaching. It's not just an AI experiment; it's a living, reflective portfolio and professional coach. Everyone always says that the big thing AI fails to do is to create reflective responses, but to a certain extent, with MechaBeth, I have managed it.
This blog post tells the story of how MechaBeth came to be, what it’s been used for, the unexpected impact it’s had, and most importantly how you can build something similar for yourself.
The Inspiration: Simulated Patients
It all started with a simple idea: could I create a virtual patient to help dental students practise communication, empathy and safeguarding scenarios? As someone deeply involved in simulation-based education, this has always seemed like the next stage to immersive clinical training. Hiring actors is expensive and using tutors to simulate patients is time consuming, so I wondered if I could create a prompt that would enable an effective discussion with my students. I initially used the “act as” feature of ChatGPT, using a prompt to get the LLM to act as a patient with specific dental needs, here is the prompt:
“I want you to act as a very nervous patient about to have a root canal. You are unaware of what the treatment entails and will question every aspect of it, you are also needle phobic and do not want an injection. You will want to ask at different stages of the conversation why you require a root canal, what the risks and benefits are and what the procedure involves. I will act as the dental nurse, you will act in a way that will encourage me to respond empathetically. Only agree to the treatment when you think I have lessened your fears significantly”
(Just to add, I have to put that final line in there, otherwise I quickly realised it would not stop questioning!)
Interestingly, I have now used this exercise twice with my students; a year ago, students were outraged that I was getting them to use ChatGPT in a teaching session! At the time, ChatGPT wasn’t really featured in teaching, it was still lurking in the shadows only regarded as that villainous tool that enables students cheat essays. I think my students held a belief that I was tricking them; if I saw them with ChatGPT on their phones, I could use that as evidence to suggest they had had used it in their writing. Easy step to fix that - ditch the essays and opt for more authentic and secure assessments, but I digress. The students opted to do the task as a group instead whilst I logged in on my on account on screen. They enjoyed it, but we lost the opportunity to see the range of responses, and compile what we thought what would be the most effective in that scenario. Fast forward a year and the students are happily embracing ChatGPT and were eager to do the task. They could all have their own conversation, export the chats, and swap them around so we could evaluate each others communication techniques, and most importantly, use the resource (or prompting technique) to practice at home.
The Outcome: A Simulated Practitioner
As I was experimenting with responses to my own virtual patient prompts, I moved to using ChatGPT’s custom GPT feature, so I wouldn’t need to keep pasting in the prompt, I could have a different chatbot for every patient scenario. In order to train a custom GPT, you need to provide it with as much background information as possible to effectively train it. In order to test it, I created one with information I had, my own. I did remove any sensitive or personal data, but I added in things like CPD reflections and my CV. After playing with it for some time I realised that using my own data was incredibly useful, just not necessarily for my original plan of a fictional patient.
I named it MechaBeth after a school nickname, Megabeth, and I turned the idea on its head; what if the simulated patient became the simulated practitioner? I started asking it questions about my career to test how much of the training information I provided had been embedding and to beadily look out for possible hallucinations. I put my own job description into ChatGPT and asked it to generate 50 questions based on it. I then fed those questions back into the chatbot which gave insightful, experience-led answers which I found fascinating. I would try to answer the question to the best of my ability at first, and then watch MechaBeth spew forth a more concise answer, with no hesitations and sometimes linking to past expereince I had forgotten to mention. Not only did I see this as profoundly useful as interview expereince for students, but it also instilled a new confidence in myself, it helped remove that imposter syndrome and reminded me that I was good enough, that I had the answers to many difficult questions, I just needed to be able to respond like the chatbot did.
Being able to confidently articulate answers to interview questions under pressure is a different skill entirely, and I knew that is exaclty what I needed MechaBeth to do to take it to the next level (and help in a job interview for a Learning Designer role!). What I did next, was to take the questions and responses and put them into ElevenLabs. ElevenLabs is an AI company that specialises in realistic voice synthesis, offering tools like text-to-speech, voice cloning, and multilingual AI dubbing. Their technology is widely used in audiobooks, video narration, and accessibility, with ethical safeguards in place to prevent misuse. I used the text-to-speech generator to generate the questions in one voice, and the responses in another. Then I edited it all together into a Podcast, so I could have the surreal experience of listening to myself be interviewed whilst I was running at the gym. If I wanted to be really authentic, ElevenLabs to have a very cool voice cloning tool which will recreate your voice after just a few phrases of input. I created mine just by reading health and safety posters out loud around the office, I then used the finished voice clone to create an AI avatar, but more on that later! Here is an excerpt of the Podcast I made:
Other uses
It’s not only interview prep that this technique could benefit, it does still serve as a basis for creation of virtual patients. Interview prep: I run potential questions through the chatbot and refine my answers using what it suggests. It can also help with application writing, particularly working as a guide to remind me of past projects I might forget or underplay. It could also form the basis of a Module help chatbot, and I can use it with students or mentees to show how to approach reflective or values-based questions. Recently, I’ve been playing with another chatbot I have created to respond as me, and I have also included my anonymised calendar as part of it’s training, so it will even respond with my availability. I’ve even added a QR code linking to MechaBeth on my CV, so employers, collaborators, or curious colleagues can ask it questions about my work, my philosophy, or even my teaching methods. You can have a go yourself here:
Use the QR code to access my MechBeth chatbot
The Limitations
It’s not perfect. There are a few drawbacks, the main one being that only people with a ChatGPT Plus subscription can use it, as it’s built on the GPT-4 model with custom instructions. I have been looking around for alternative custom chatbot tools that are free such as Landbot, but most are only free until you reach a certain number of interactions which would not be viable for continuous student use. It’s also worth noting that over-reliance is a big issue; did it quell my interview fears? No, I was a nervous mess, and that was possibly down to over-preparation, obsessing over answers to questions that were never even asked. But did I get the job? Yes. It would be interesting to study how effective it is with students, getting them to do a mock interview on their own and then see how they fair with it. At the end of the day though, it is a tool for reflection, not a replacement for your voice, and it’s vital to still use it critically. There are also the obvious privacy concerns, you must ensure the documents you train it on don’t contain patient data or sensitive personal information. For personal use, if you are willing to pay the ChatGPT Pro subscription, the benefits have far outweighed the limitations. And as a bonus, it’s a great way to demonstrate AI literacy, innovation, and reflective practice, qualities that are increasingly valued in higher education and health care settings.
And where does this leave virtual patients? I still feel the best way currently is using the “act as” prompt. You can still use AI to generate images or case studies for your patients, but the difficulty is to match up the Large Language Model response with a live video. We can create AI Avatars, but they work on pre-written text, theres no live generation (yet). As I eagerly await the advancement of AI Avatars, perhaps this could work in the meantime as some sort of branching scenario tool, but it would be a lengthy process. These AI Avatar tools are also very pricey, but they do look impressive. Here is one I made using a free trial of D:ID, a photograph of myself, and my Elevenlabs voice clone:
How to build it
Want to create your own version of MechaBeth? Here’s how to do it using ChatGPT’s custom GPT tools:
Step 1: Audit Your Content
Before anything technical, start by gathering the material that best represents you:
CVs
Portfolios
Teaching observations
Personal reflection
PDPs
Example interview answers
CPD logs
Publications or research presentations
(Make sure there’s no confidential data, nothing that isn’t already public knowledge)
Step 2: Access ChatGPT Custom GPTs
Get a ChatGPT Plus subscription.
Go to https://chat.openai.com.
Click Explore GPTs or navigate to Create a GPT under your profile.
Select Create a GPT, and follow the setup prompts.
Step 3: Define the Personality and Goals
When asked to describe your GPT, you can write something like:
“This GPT acts as a virtual version of (your name here). It answers questions as if it were (name), using the CV, CPD records, reflections, and teaching experience provided. It specialises in interview preparation and uses specific examples from their career to illustrate points.”
Step 4: Upload Documents
Upload your selected files directly during the setup process, this can include:
Word documents
PDFs
PPT slides
Annotated CPD logs
These will be ingested by the model and used to inform answers. You don’t need to format them; just make sure they’re accurate and representative.
Step 5: Add Custom Instructions
You can tailor how it behaves by giving it explicit instructions, like:
"Always respond in my voice: professional, supportive, reflective."
"Use real examples from my portfolio to answer questions when appropriate."
"Avoid guessing or using experiences that aren’t documented in the uploaded material."
Step 6: Test and Refine
This is the step to have some fun with it though, and you do get an example of the output displaying on the right hand panel so you can test it out, I recommend tweaking the instuctions until it accurately reflects what you want. Ask it a few sample questions:
“How would you respond to a question about interprofessional collaboration?”
“What is your teaching philosophy?”
“Give me an example of how you incorporated NHS values into dental education?” (you can even upload the values with the question to be specific)
Step 7: Share It
You’ll get a unique link to your GPT. You can:
Add it as a QR code at the bottom of your CV.
Use it as an example of innovation.
Share it in workshops on AI literacy or reflective practice.
A really useful exercise it to share it with those that know you, see how much they think it represents you, especially as your own opinion is likely to be biased.
Final Reflections
Building MechaBeth has changed how I engage with my own story. It’s helped me see connections I’d missed, gain clarity in interviews, and share my journey in a way that’s structured but deeply personal. It also reflects how far educational technology has come, from simulation blocks to simulated practitioners. If you’re someone with a hybrid or interdisciplinary career, or someone who struggles with imposter syndrome, I can’t recommend this process enough. It’s not just a chatbot. It’s a mirror, a practice partner, and a personal archive. You don’t need to be a tech expert to build it. You just need a story worth telling and trust me, you already have that. If you still don’t believe me, your chatbot will prove it.
