Hive Hub

The problem

Keyword search is brittle when a candidate and an employer describe the same ability differently. Hive Hub explores whether semantic retrieval can make those near-matches visible without hiding the underlying job information.

What I built

The supplied resume records a LangChain recommendation flow, Prisma, Pinecone, Google Gemini embeddings, and PostgreSQL with pgvector. The public repository also contains authentication, rate limiting, and email integration.

The interface keeps the recommendation beside the original job details. A score is a navigation aid, not a verdict about a person.

The systems path

candidate profile -> normalized text -> embedding
job description   -> normalized text -> embedding
                  -> similarity search -> ranked explanation

What I learned

Vector similarity is only one input to a useful recommendation. A trustworthy product must explain why a result appeared, preserve filters the user controls, and make it easy to inspect the source job.