Matchly is a full-stack AI-powered resume intelligence platform that analyzes resumes and job descriptions, generates tailored cover letters, and provides insightful feedback. Built with Next.js, FastAPI, and containerized microservices.

• Frontend: Next.js with shadcn/ui
• Backend: FastAPI
• Vector Database: ChromaDB
• Containerization: Docker & Docker Compose
• Styling: Tailwind CSS, Google Fonts (Geist)
• PDF Parsing: PyMuPDF (via fitz)
• LangChain: Prompt templating & chaining
• Ollama (Mistral): Local LLM backend

Matchly/
├── frontend/           # Next.js frontend app
│   ├── app/            # App router and layout logic
│   ├── components/     # UI components (ResumeUploader, JDInput, etc.)
│   ├── lib/
│   │   └── api.ts      # Frontend API interaction with FastAPI
│   └── ui/             # shadcn/ui components
│
├── backend/            # FastAPI backend
│   ├── main.py         # Application entrypoint
│   ├── resume_utils/   # Resume parsing, embedding, matching logic
│   │   ├── parser.py   # PyMuPDF-based PDF text extraction
│   │   ├── embedder.py # Text embedding for vector comparison
│   │   └── matcher.py  # Matching algorithm between resume and JD
│   └── api/            # FastAPI route handlers
│       └── process.py  # /api/process endpoint
│
├── docker-compose.yml  # Orchestrates containers
└── README.md            # Project documentation

• File upload interface for resumes and job descriptions
• Input fields for job title and hiring manager name
• Editable text area for the generated cover letter
• Copy to clipboard and download as PDF functionality for the cover letter
• Handles FormData submission to FastAPI backend
• Responsive UI built with Tailwind CSS and shadcn/ui components

• Resume and JD parsing with PyMuPDF (fitz)
• Embedding and similarity scoring using ChromaDB
• Cover letter generation via LangChain-prompted Mistral LLM (local via Ollama)
• Supports RAG, prompt chaining, tone control, and personalized output
• REST API exposed at /api/process with CORS enabled

• Local LLM Inference: Integrated Mistral via Ollama for fast, private, and offline cover letter generation.
• LangChain Prompt Chaining: Modular prompt templates and chaining logic enable role-specific, tone-adjustable letters.
• RAG Workflow with ChromaDB: Combines resume and job description context via a Retrieval-Augmented Generation (RAG) setup, using ChromaDB for fast vector similarity search.
• Semantic Personalization: Prompts adapt based on parsed resume, job title, and hiring manager details for contextual generation.
• Editable AI Output: LLM-generated cover letters are editable in the UI to allow manual refinement and customization.

• Dockerized setup for frontend, backend, and ChromaDB
• Local development using Docker Compose
• Environment supports hot reload for both FastAPI and Next.js
• Easy extension for cloud deployment (Render / Railway / Vercel)

This project uses Docker Compose to run the full stack: Next.js frontend + FastAPI backend + ChromaDB vector store.

git clone https://github.com/sourabhaprasad/matchly.git
cd matchly

docker-compose up --build

This will:

• Serve the Next.js frontend on http://localhost:3000
• Start the FastAPI backend on http://localhost:8001
• Launch the ChromaDB vector database

If you modify code or dependencies:

docker-compose down
docker-compose up --build

Accepts multipart FormData and returns match results and a generated cover letter.

{
  "match_score": 0.84,
  "matched_skills": ["Python", "Machine Learning", "Data Analysis"],
  "uunmatched_skills": ["Kubernetes", "AWS Lambda"],
  "cover_letter": "Dear Hiring Manager,\nI am excited to apply for the role of..."
}

ChromaDB is used to store and compare semantic embeddings between resumes and job descriptions.

• Persistence: Stores vector data locally at ./chroma_db/
• In-Process: No separate service required; runs with FastAPI

To reset embeddings:

rm -rf backend/chroma_db/

• Integrate GPT-based or OSS LLMs for:

  • Enhanced resume and JD understanding
  • Dynamic and role-specific cover letter generation

• Improve scoring with custom similarity metrics and weights

• Add user authentication (via Supabase or NextAuth)

• Store upload history, cover letters, and match scores per user

• Production deployment with CI/CD pipelines

• Install a new shadcn/ui component:

  npx shadcn@latest add <component>

• Add a frontend dependency without rebuilding the container:

  docker exec -it <frontend-container-id> sh
  npm add <package-name>

• Common Issues:

  • Cannot find module 'sonner'

    • Run: npm add sonner inside the frontend directory

  • CORS errors on frontend

    • Ensure CORS middleware is enabled in main.py in FastAPI backend

• Sourabha Prasad – GitHub