How chandrima0503/ats-ai works
Project Type and Purpose
• This is a full-stack, AI-powered Applicant Tracking System (ATS).
• Frontend: Vue 3 + Vite single-page application for uploading resumes, creating job postings, and viewing ranked candidates.
• Backend: Python FastAPI service that parses PDF resumes, generates embeddings, stores them in ChromaDB, and ranks resumes against job descriptions.
File Structure Analysis
• Top-level:
• backend/ – FastAPI application
– app/core/config.py (configuration)
– app/main.py (entry point, mounts routes)
– app/routes/{jobs.py, match.py, resumes.py} (API endpoints)
– app/services/{parser.py, embedder.py, gpt.py} (business logic)
– app/models/schemas.py (Pydantic schemas)
– data/vector_store/chroma.sqlite3 (persistent vector DB)
• frontend/ – Vue 3 app
– src/components, src/views, src/router, src/stores, src/composables/useApi.ts
– Cypress tests, Vite + TypeScript configs
• CI: .github/workflows/ci.yml
• Deployment: Dockerfile in backend/
Key Architectural Insights from README
• Resume upload → PDF parsing (pdfplumber + spaCy)
• Text embedding (sentence-transformers)
• Vector store in ChromaDB (SQLite persistence)
• Job descriptions embedded similarly
• Ranking resumes by vector similarity
• All operations exposed via FastAPI REST endpoints
a. Main Components
• Client Browser (SPA)
• API Gateway / FastAPI Backend
• PDF Parser Service (pdfplumber + spaCy)
• Embedding Service (Sentence Transformers + optional GPT)
• Vector Store (ChromaDB / SQLite)
• Frontend Unit & E2E Test Harness (Cypress)
• External Services (if any: e.g., model hosting or OpenAI API)
b. Relationships / Data Flow
1. User uploads PDF → HTTP POST to /resumes/upload
2. FastAPI controller invokes Parser Service → extracted text & metadata
3. Text passed to Embedder Service → vector generated
4. Vector + metadata stored in ChromaDB (SQLite file)
5. User submits job description → similar pipeline (embed + store)
6. Ranking request → Backend queries ChromaDB for nearest vectors → returns ranked results
7. Frontend displays results
c. Architectural Patterns
• Layered architecture: Controllers (routes) → Services → Data Store
• Separation of concerns: Parsing, embedding, storage, API routing
• Persistent vector store (ChromaDB) for fast similarity queries
d. Technologies / Frameworks
• Frontend: Vue 3, Vite, Vue Router, Pinia, Axios (useApi.ts), Cypress
• Backend: Python 3.11, FastAPI, uvicorn, Pydantic, spaCy, pdfplumber, sentence-transformers, chromadb
• Deployment: Docker
• CI: GitHub Actions