01 / RELATIVE TECH SUPPORT
Relative Tech Support.
A patient AI tech helper for the moments when your technology refuses to cooperate. Describe the problem, share an image if it helps, and work through it conversationally, one step at a time.
The app opens on Hugging Face. This is a functioning experiment that I’m continuing to improve.
LIVE APP · IN PROGRESS
THE STARTING POINT
A familiar question.
I’m one of the people my family calls when something stops working. Those conversations usually start with a question, a little patience, and figuring out what the person is actually seeing.
While taking an LLM engineering course, I wanted to make something of my own alongside the exercises. That familiar role suggested a small experiment: could I make an AI assistant that approached troubleshooting the same way?
The first version was a simple chatbot with a helpful-grandson personality. I could try it out and see whether its explanations made sense to someone who didn’t know the technical terms. How it asked questions mattered as much as the answers it gave.
FROM CHAT TO APPLICATION
Each addition raised a new question.
- 01
Working chat
Get the conversation working, then refine the instructions so the helper asks useful questions and gives manageable steps.
- 02
Image support
Let someone show an error or a screen. Work out how that image belongs in the conversation, alongside what they’ve already said.
- 03
Improved structure
Move from notebook experimentation into an application with clearer responsibilities, configuration, and a Git history I can return to.
- 04
Live application
Get it running on Hugging Face and consider what changes when someone else can use it: secrets, access, security, and API costs.
WHAT STUCK WITH ME
The chat was only part of it.
Patience needs to show up in the response.
A friendly personality isn’t enough. Asking one useful question and offering one understandable next step matters more than giving someone a long list of possibilities.
A conversation has a shape.
Remembering earlier messages, including an image, and deciding what the model receives made conversation handling a real part of the application rather than an invisible detail.
Then it left my computer.
Putting RTS online meant someone else could use it without my setup, my machine, or me sitting there to help. It also meant I couldn't control what someone might ask it or send to it.
That brought a different set of concerns into the project: deployment and configuration, keeping secrets out of the code, deciding how people could access it, thinking about security, and understanding what public API usage could cost.
Made with Python, Gradio, and OpenAI. Hosted on Hugging Face, with the source and deeper implementation notes on GitHub.