Walk Tutor: a voice DSA coach that runs offline on my laptop

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Walk Tutor is a voice coach for practicing data structures and algorithms (DSA) without staring at a screen. It reads you a problem, you explain your approach out loud, and it answers with a hint or a follow-up question. You can say "next problem", "repeat" or "quit", so there is no keyboard at all once it starts.

I'm a third-year student, and most of my practice is sitting at a laptop. I built this so I could practice while walking outside instead.

Code

ramya2006sri / walk-tutor

Walk Tutor

A voice DSA coach you use on a walk. It runs fully offline on a laptop.

How it works

• Speech to text: faster-whisper
• Coach: Gemma 3 1B through Ollama
• Speech out: pyttsx3
• The code checks your answer for the key idea; the model only writes the feedback.

Setup

1. Install Ollama and run: ollama pull gemma3:1b
2. pip install -r requirements.txt
3. python walk_tutor.py

Voice commands: "next problem", "repeat", "quit".

View on GitHub

How I Built It

Everything runs locally on my Windows laptop:

• Gemma 3 1B through Ollama writes the coach's replies.
• faster-whisper (open-source Whisper) turns my speech into text.
• pyttsx3 speaks the answers using the voices built into Windows.
• A small Python loop detects when I start and stop talking, so it needs no button presses.

The lesson that shaped the design: my first version let Gemma judge my answers. The 1B model was far too agreeable. I told it I'd solve a bracket-matching problem by reversing a linked list, and it called that a great approach.

So I split the jobs. Each problem has a few keywords for its key idea (for Valid Parentheses, the word "stack"). The code checks my answer for them and decides the verdict. Gemma only writes the feedback, and it must follow the verdict. That made the small model reliable. The tradeoff is that a keyword check is simple. Saying the right word in a wrong solution can still pass.

Other things I fixed while testing: it cut me off during long answers, so I raised the pause time and the answer limit.

Why Does Open Innovation Matter?

• It works offline. After the one-time model downloads, it needs no internet, which matters on a trail with no signal.
• My voice stays on my laptop. Nothing is sent to a server I don't control.
• It costs nothing to run. No API key and no usage bill.
• I could change how it behaves. I rewrote the coach's rules and added the code-checked verdict myself. With a closed chat API, I wouldn't have been able to work around the model's weaknesses this way.

Prize Categories

Best Use of Gemma