AI-Powered Semantic Job Matching System Using FastAPI, Vector Databases, and Dual Encoders

Short summary

Developer built JobSync, a semantic job-matching system using vector embeddings and FastAPI. Project compared vector databases (Qdrant vs pgvector), explored remote LoRA fine-tuning, and revealed that production AI challenges are less about models and more about infrastructure, deployment, and system design.

• •Built semantic matching system using dual-encoder architecture and vector embeddings instead of keyword matching
• •Compared Qdrant (HNSW) and pgvector (IVFFlat) for vector search performance—Qdrant showed significantly faster retrieval
• •Learned production AI is about infrastructure, APIs, and deployment challenges—not just model training

Generated with AI, which can make mistakes.