PrepAI: An Offline AI Interview Coach Powered by Open-Source AI

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

Preparing for technical interviews can be difficult, especially for students who don't have access to mentors or enough opportunities to practice.

Many AI interview preparation tools depend on cloud-based APIs, which can introduce recurring costs and require users to send their answers to external services.

I built PrepAI, an AI Interview Coach that helps students practice interviews and improve their answers using a locally running AI model.

PrepAI simulates an interview by asking questions, evaluating responses, and providing actionable feedback.

Demo

Video -

prepai video.mp4 - Google Drive

drive.google.com

Code

https://github.com/danish16102005-hub/Ai-Interview-Coach

How I Built It

I built PrepAI using React for the interface and Google's Agent Development Kit to manage the interview conversation.

Technology Stack
Technology Purpose
React Build the interview interface
Vite Frontend development and build tooling
Python Implement the AI agent
Google ADK Manage the agent's instructions, tools, and conversation
Ollama Run the AI model locally
Qwen 2.5 3B Power the interview agent
LiteLLM Connect ADK to the local model

Why Does Open Innovation Matter?

Open innovation matters because useful AI tools should not always require expensive infrastructure or proprietary APIs.

For PrepAI, open-source and open-weight technologies made it possible to combine an agent framework with a locally running language model.

This has several potential benefits:

1. Greater control over data

When inference runs locally, interview answers can remain on the user's computer, provided the application does not send them elsewhere.

1. Reduced dependence on paid APIs

Users do not need to pay a cloud provider for each model request when using local inference. They still need suitable hardware and must account for electricity and setup costs.

1. Flexibility

Developers can experiment with different compatible models and improve the agent's instructions and evaluation tools without redesigning the entire application.

1. Learning and experimentation

Using an open agent framework and a local model makes it easier to understand how the system works instead of treating AI as a black box.

For me, the biggest benefit is being able to build an AI-powered application while learning how agent frameworks, tools, and local inference work together.

My Agent Session

Prize Categories

Best Use of GitHub Copilot