I Built a Mock Interviewer for Friend That Never Leaves My Laptop

Title: I Built a Mock Interviewer for My Friend That Never Leaves My Laptop

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

My friend Anushka is a B.Tech CSE graduate preparing for off-campus placements. She told me the part she dreads most isn't the technical round. It's the silence right after "Tell me about yourself."

"No matter how many exams I qualify, I always mess up in the interview round and end up with no response."

That's when I realised how many students have the same problem: they can practise coding, but they have nobody to practise being interviewed with. So I built her a practice partner: Interview Buddy.

You pick the role and the round (HR, technical, or mixed). It asks one question at a time, like a real panel. After every answer you get:

• a score out of 10
• one thing that worked
• one thing to fix
• a stronger version of your own answer, not a generic template answer

At the end there's a short report: strengths, what to practise, and what to do this week.

Demo

Code

https://github.com/Vaishnavishrivastav/Interscene

How I Built It

The whole thing is small on purpose:

• Gemma (open-weight, by Google) runs locally through Ollama. No API key, no account.
• Flask serves one page and one endpoint.
• The server is stateless. The browser sends the conversation history on every turn.
• The model replies in a strict format (SCORE, WORKED, IMPROVE, STRONGER, NEXT QUESTION), and the frontend parses it with a forgiving regex, because small models don't always follow formats perfectly.

Swapping models is one environment variable:

OLLAMA_MODEL=gemma3:12b python app.py

Small model for a weak laptop, bigger one for better feedback.

Why Does Open Innovation Matter?

Think about what someone says in a mock interview. Their weaknesses, gaps in their projects, nervous half-formed answers they'd never say to a stranger.

I didn't want any of that on a server she doesn't control. Because Gemma runs on the laptop, her answers never leave it, it costs nothing per message, and it works with the Wi-Fi off. She can fail badly at question one, try again, and nobody sees it.

A closed API would have worked technically. But "your answers never leave this machine" is a feature she can actually feel, and it only exists because the model is open and local. Swapping the model with one line is the other thing a hosted API wouldn't give me.

What She Said

"It was truly very helpful to me for my preparation." — Anushka

Prize Categories

• Best Use of Gemma: Gemma runs locally via Ollama and powers every question, score and piece of feedback.