Inspiration

I'm a first-semester electronics student. I learn by practising, but chatbots can't be trusted to mark numeric answers. I wanted a tutor where the maths is always exact. I also wanted to build it on my phone, because I don't own a laptop.

What it does

Electronics Tutor is a phone-friendly web app with two tabs.

Ask: answers electronics questions in simple English. A second AI call reviews each answer and shows the result in a "Second check" box.
Quiz: practice questions on Ohm's law, series resistors, parallel resistors and power. After a wrong answer, it shows the correct working and explains the likely mistake.

How I built it

Python makes each quiz question from random numbers and works out the correct answer itself. It checks your number with a small tolerance, so marking never depends on an AI's opinion.
The AI only explains. After a wrong answer, it is given the correct value and describes the likely mistake.
The interface is Gradio. The AI is NVIDIA Nemotron models, called through the NVIDIA Build API.
It runs in Google Colab, and the code is on GitHub. API keys and the app password are typed into hidden prompts, so no secret is stored in the code.

Challenges I ran into

I had no laptop, so I built everything on my phone, including fixing indentation errors from copied code.
My first version let the AI grade answers, which was unreliable. Moving the checking into Python fixed it.
Nebius Token Factory needs a payment card to finish setup, which I couldn't complete. This version therefore runs on the NVIDIA Build API, not on Nebius. Moving over means changing the web address, the key and the model names.
Partway through, my model name stopped working with a 404 error. I learned to list the models my key can use and test which ones answer.

Accomplishments that I'm proud of

A working tutor with exact, Python-checked quizzes, built with no prior hackathon experience.
A protected app, a public GitHub repo, and a README that honestly lists what it can't do yet.

What I learned

Prompts are part of the program.
Use code where answers are exact, and AI where they aren't.
A second AI reviewer helps, but it can share the first one's blind spots, so a different model gives a better check.
How APIs, keys and error codes like 401 and 404 work.