Job matching that reads like a recruiter, at machine scale

Challenge

A recruitment-technology client needed candidates matched to postings with the judgment of a senior recruiter — across thousands of postings from multiple sources, updated continuously, without a recruiter in the loop for the first pass.

Approach

We built a full-stack matching platform: multi-source ingestion feeds a hybrid retrieval layer that combines BGE-M3 dense embeddings with BM25, fused by Reciprocal Rank Fusion. A cross-encoder re-ranks the shortlist, and an LLM scores final fit with structured reasoning. A React dashboard gives operators control; automated unit, integration, and end-to-end tests guard every release.

Architecture

Results

• —Recruiter-grade shortlists from a fully automated first pass
• —Production AWS architecture with infrastructure as code and one-command deploys
• —Test coverage across the stack: unit, integration, and E2E suites in CI