BreadWinner AI 🍞

Developed at ConUHacks X (2026)

BreadWinner AI is an advanced, AI-driven interview simulation and resume optimization platform designed to bridge the gap between academic learning and the rigorous demands of technical interviews. Built during a 24-hour sprint at Concordia University, it empowers students to practice with realistic AI avatars, optimize their resumes, and track their job search progress in one cohesive ecosystem.

🚀 Project Mission

Technical interviews are a distinct skill set from software engineering. Many qualified students fail not because of a lack of technical knowledge, but due to a lack of preparation for the interview format itself. BreadWinner democratizes access to high-quality mock interviews, providing real-time feedback and personalized study plans to help candidates land their dream jobs.

✨ Key Features

• 🤖 AI Interview Simulations: Real-time, voice-interactive video interviews with an interactive AI avatar (powered by HeyGen & OpenRouter). The AI adapts questions based on your responses and provides instant feedback.
• 📄 Smart Resume Parsing: Upload PDF/DOCX resumes to automatically extract structured data using PyMuPDF and python-docx, enabling gap analysis against job descriptions.
• 🔄 Automated Portfolio Updates: Seamlessly syncs resume data with your user profile to keep your professional identity up-to-date.
• 📊 Job Application Tracking: A Kanban-style dashboard to manage applications, track status, and view "Match Scores" for specific roles.
• 🧠 Study Lab: Generates personalized technical, behavioral, and LeetCode study materials based on the specific requirements of a target job.

🛠️ Tech Stack

Category	Technologies
Backend	FastAPI (Python 3.11), Pydantic, SQLAlchemy
Database	PostgreSQL (Hosted on Neon), Psycopg2
Frontend	React (Vite), TypeScript, Tailwind CSS, Lucide Icons
AI & ML	OpenRouter (Gemini 2.0/GPT-4), HeyGen SDK, Firecrawl
Infrastructure	AWS Lambda, Amazon ECR, AWS Amplify, Docker
Tools	Git, Postman, Docker Compose

🏗️ Architecture & Infrastructure

We adopted a Serverless Architecture to ensure scalability and cost-efficiency, crucial for a hackathon project with potential for high burst traffic.

Backend: AWS Lambda & API Gateway

• Containerization: The FastAPI backend is packaged as a Docker container. We utilized a multi-stage build process to ensure binary compatibility with Amazon Linux 2023, specifically handling glibc version mismatches for libraries like PyMuPDF.
• Mangum Adapter: We used Mangum as an adapter to wrap our ASGI FastAPI application, allowing it to process events from AWS Lambda and API Gateway seamlessly.
• Proxy Integration: API Gateway is configured with a proxy integration, forwarding all requests to the Lambda function. This allows FastAPI to handle routing internally, simplifying the cloud configuration.

Frontend: AWS Amplify

• The React frontend is deployed via AWS Amplify, providing a robust CI/CD pipeline that automatically builds and deploys changes from our GitHub repository.
• Environment variables are managed within Amplify to point to the dynamic API Gateway endpoints.

🧠 Lessons Learned

Building a full-stack AI application in 24 hours came with significant challenges. Here are our key takeaways:

1. Serverless Constraints & Docker

Deploying Python applications with C-extensions (like psycopg2 and PyMuPDF) to AWS Lambda is non-trivial. We learned that local Docker environments (Debian-based) differ from Lambda's runtime (Amazon Linux).

• Solution: We switched to the python:3.11-slim-bullseye base image and forced pip to use manylinux2014 wheels to ensure binary compatibility.

2. The CORS Conundrum

Connecting a frontend hosted on one domain (Amplify) to a backend on another (API Gateway) triggered strict CORS policies.

• Solution: We configured FastAPI's CORSMiddleware to accept specific origins dynamically and ensured our API Gateway responses included the necessary Access-Control-Allow-Origin headers.

3. Stateless Authentication

Since Lambda functions are ephemeral, storing session data in memory (like a simple token dictionary) caused failures when requests hit different container instances.

• Solution: We implemented stateless authentication using JWT (JSON Web Tokens), allowing any Lambda instance to verify user identity without shared state.

📦 Installation & Setup

Prerequisites

• Python 3.11+
• Node.js & npm
• Docker (for deployment)

Backend Setup

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Create .env file
cp .env.example .env

# Run locally
uvicorn main:app --reload

Create a virtual environment and install dependencies:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

Create a .env file in backend/ with the following variables:

DATABASE_URL=postgresql://user:pass@host/dbname?sslmode=require
OPENROUTER_API_KEY=your_key
GEMINI_API_KEY=your_key
HEYGEN_API_KEY=your_key
FIRECRAWL_API_KEY=your_key
JWT_SECRET=your_secret
ALLOWED_ORIGINS=*

Run the local server:

uvicorn main:app --reload

2. Frontend Setup

Navigate to the frontend directory:

cd frontend/interviewPrep

Install dependencies:

npm install

Create a .env file in frontend/interviewPrep/ (or .env.local):

VITE_API_URL=http://localhost:8000

Run the development server:

npm run dev

☁️ Deployment

Backend (AWS Lambda)

The backend is containerized and deployed to AWS Lambda using a custom Docker image.

1. Build the Docker image:

   docker build --platform linux/amd64 -f backend/Dockerfile -t breadwinner-backend backend

2. Push to AWS ECR:

   aws ecr get-login-password | docker login --username AWS --password-stdin <your-account-id>.dkr.ecr.<region>.amazonaws.com
   docker tag breadwinner-backend:latest <your-repo-uri>:latest
   docker push <your-repo-uri>:latest

3. Update Lambda: Update the Lambda function to use the new image URI.

Frontend (AWS Amplify)

1. Connect your repository to AWS Amplify.
2. Set the environment variable VITE_API_URL in the Amplify console to your Lambda Function URL or API Gateway endpoint.
3. Trigger a build.