Embedding-Based Job & Candidate Matching Platform (Vector Search + FastAPI + React)

Kaloina Melodie Ravoahangilalao
Published

0
  • Project creator

Traditional recruitment platforms rely on keywords. But keywords do not capture meaning.

I re-engineered an existing job-matching system by introducing an embedding-based semantic search workflow to improve accuracy and scalability.

The core idea: encode both CVs and job offers as vector embeddings and perform similarity search in Milvus instead of running real-time LLM inference for every request.

The pipeline includes:

CV parsing with LLM-based structured extraction (PDF/DOCX → JSON)

Embedding generation and storage in Milvus

FastAPI backend exposing matching and search endpoints

React frontend for recruiters and candidates

Dockerized monorepo with CI/CD via GitHub Actions

Deployment on Render (frontend) and Google Cloud Run (backend)

By encoding documents once and reusing vectors, the system significantly reduces latency and improves scalability as the dataset grows.

This project reflects strong understanding of semantic search, vector databases, API architecture, and production-grade deployment.