I built an AI assistant
I woke up at 6:30am on a Tuesday with no code written and no product built.
By lunch, I had a working AI assistant that listens to my voice and adds things to my Google Calendar automatically.
This is Week 1 of the 10x10 Build Challenge — me and Robin Hagqvist are each building 10 products in 10 weeks. Every week, one new thing. Ship it, learn from it, write about it.
Here's what happened this week.
The problem
I have a lot going on. University, VES, side projects, friends, gym. I have 14 Google Calendars. And every time I need to add something, I have to unlock my phone, find the right calendar, type everything in manually.
It's not hard. But it's annoying enough that I sometimes just don't do it.
I wanted something I could just talk to.
What I built
A Telegram bot — @durppppp_bot — that takes voice or text messages and adds things to Google Calendar automatically.
You say "add a gym session tomorrow" and it finds a free evening slot and puts it in your Gym calendar. You say "meeting with Mikael on Friday at 3pm" and it checks for conflicts first. You say "mitä minulla on tänään?" and it reads back your day in Finnish.
It also messages you every morning at 8am with your schedule for the day.
Total cost: €0. Total build time: one morning.
What I used
Telegram Bot API — free, no approval needed
Groq Whisper — free voice transcription, works in Finnish too
LLaMA 3.3 — free AI to understand what you said
Google Calendar API — reads and writes your calendar
Railway — hosts it 24/7 for free
Python — about 400 lines total
What surprised me
I expected the AI to be the hard part. It wasn't. You give it a message, it gives you back structured data. Done.
The hard parts were boring: Google OAuth setup, stopping the bot from running twice at the same time, making sure credentials don't expire on the server. Classic infrastructure stuff nobody talks about.
Finnish also worked instantly. Whisper transcribed "Mitä minulla on tänään?" perfectly on the first try. I didn't do anything special for that.
What I learned
The AI part of "AI products" is actually the easy part now. The integrations are what take time.
Also: deciding exactly what to build before writing any code saved me hours. We spent 30 minutes scoping before touching the keyboard. That 30 minutes was worth more than any single hour of coding.
What's next
Week 2 starts Monday. Different theme, different product.
If you're a student who wants to build things but doesn't know where to start, one morning is genuinely enough to ship something real. You don't need a team, funding, or months of planning.
You just need to start.
10x10 Build Challenge — Week 1 of 10 Built by Anton Heinonen alongside Robin Hagqvist
