I built my
own Jarvis.
I wanted to talk to my laptop like Tony Stark talks to Jarvis: ask a question, set a timer, get a study plan written up, without stopping what I'm doing. Siri couldn't do most of that, so I made my own and called it Alex.
You say "Hey Alex". It listens, thinks using an AI model on my own server, and answers out loud in a British accent. It opens apps, plays music, tracks how long I've studied for each course, and writes Word and PDF reports straight into my Documents folder.
Built with Python, speech recognition, an ElevenLabs voice and Claude AI models. The idea, the features, the testing and every decision were mine, and I directed AI coding tools to write the code.
it pulses when Alex talks
180 minutes that day
It woke up for no reason
I tested the wake word against 14 phrases that sound close, like "Alexa" and "Hey Alice". After tuning, it ignored all of them.
It was slow
Every question went to the biggest AI model. Now simple ones go to a faster model and only heavy tasks use the big one. Replies take about half the time.
You couldn't shut it up
If it misheard me, I had to wait for it to finish talking. I added a way to cut it off mid-sentence.
I tried getting Alex to answer my WhatsApp calls and take messages when I'm busy. It could pick up, but the caller couldn't hear it, because macOS won't let another app's audio into a WhatsApp call. I parked it instead of pretending it works.
I also run my own server.
Alex's brain lives on a small Linux server I rent in Europe. I set it up to do three jobs at once, instead of paying for three separate subscriptions.
Right now
- Studying
- First semester of Business Administration at Pan-Atlantic University.
- Looking for
- An internship, part-time or remote role in operations, marketing or research, anywhere a bit of tech can save people time.
- Building
- New features for Alex, one at a time.
- Good at
- Research, reports, presentations, using AI properly, and not giving up on a problem.