reinaissanced/kudos

📄 Kudos

An AI-powered resume builder and analysis tool built with Next.js and FastAPI. Kudos uses sentence-transformers for semantic resume-to-job description matching and Gemini 1.5 Flash for AI-powered suggestions. This project was built to explore integrating machine learning models with a modern React frontend to provide instant, actionable insights for job seekers.

📦 Technologies

• —Next.js
• —React.js
• —TypeScript
• —Tailwind CSS
• —FastAPI
• —Python
• —Google Gemini 1.5 Flash
• —Sentence Transformers
• —Pytest

🦄 Features

Here's what you can do with Kudos:

• —AI-Powered Analysis: Upload your PDF resume and paste a job description to get an instant match score and skill gap analysis.
• —Secure File Handling: Process documents securely with PyPDF2 and PDF-lib.
• —Smart Suggestions: Get generative AI suggestions on how to improve your resume to pass ATS filters and stand out to recruiters.
• —Shareable Links: Generate auto-expiring shared links to showcase your analysis results or get feedback.
• —Performance & Constraints: Includes in-memory LRU caching, input truncation, and per-IP rate limiting to ensure reliability and cost control.

👩🏽‍🍳 The Process

I started by setting up the FastAPI backend, integrating sentence-transformers for calculating semantic similarities between resume text and job descriptions. Then, I connected Gemini 1.5 Flash to provide more nuanced, generative feedback on the user's skills.

Next, I focused on building the frontend using Next.js 14 and Tailwind CSS. I created intuitive components for dragging and dropping resumes, pasting job descriptions, and viewing the detailed analysis results.

To make sure the application was robust and scalable, I implemented rate limiting, input size limits, and robust error handling with exponential backoff for the AI calls.

Finally, I wrote tests for the backend using pytest to make sure the endpoints behaved as expected under various conditions. Along the way, I documented the process and environment variables so it would be easy to set up and run.

📚 What I Learned

During this project, I've picked up important skills and a better understanding of AI integrations.

🧠 Semantic Search & AI Integration:

• —NLP Models: I learned how to use sentence-transformers for local semantic matching and how to securely combine it with remote LLM APIs (Gemini) for a hybrid AI approach.

📏 Rate Limiting & Caching:

• —Performance: Implementing an in-memory LRU cache with TTL and per-IP rate limits taught me how to protect APIs from abuse and manage API costs effectively.

🎨 Next.js App Router:

• —Modern React: I deepened my understanding of Next.js, building a responsive and interactive UI that seamlessly connects with my Python backend.

🔍 Error Handling & Backoff:

• —Resilience: Writing retry logic with exponential backoff for Gemini API calls improved the stability of my app when communicating with external services.

💭 How can it be improved?

• —Add user authentication to save past resume analyses.
• —Support more document formats like DOCX or TXT files.
• —Add more granular feedback based on specific industry standards.
• —Dark mode support for the UI.
• —Use a persistent database instead of an in-memory cache for heavier loads.

🚦 Running the Project

To run the project in your local environment, follow these steps:

1. 1.Clone the repository to your local machine.
2. 2.In the backend directory, create a .env file based on .env.example and add your GEMINI_API_KEY.
3. 3.Set up the Python environment (using .venv) and run pip install -r requirements.txt.
4. 4.Start the backend server using uvicorn app.main:app --reload (or according to backend/README.md).
5. 5.Open a new terminal, navigate to the frontend directory, and run npm install or yarn.
6. 6.Run npm run dev or yarn dev to start the Next.js development server.
7. 7.Open http://localhost:3000 (or the port specified) in your web browser to view the app!

🍿 Video

(Coming soon!)