Hirelyze
AI Resume Screener
Ranks applicants with Gemini embeddings, cosine similarity and LLM-generated analysis.
system sketch
00 · system architecture
Overview
A full-stack hiring platform where recruiters post jobs and candidates apply with PDF/DOCX resumes. Applicants are ranked using Gemini embeddings with cosine similarity, plus an LLM-generated match score, strengths, weaknesses and interview suggestions.
Problem
Resume processing (text extraction, embedding and analysis) is slow, and candidates shouldn't have to wait on it when they submit an application.
Architecture
Uploads return immediately while a Redis-backed job queue hands the slow work to a background worker. Data, files and auth live in Supabase (PostgreSQL, storage and auth).
Key decisions
- 01
Asynchronous scoring
Slow resume processing is decoupled from the upload request with a Redis-backed job queue and background worker, so candidates receive an instant application ID while scoring runs asynchronously.
- 02
Embeddings plus LLM analysis
Gemini embeddings with cosine similarity rank applicants against the job, and an LLM adds a match score, strengths, weaknesses and interview suggestions.
- 03
Two audiences, two portals
A recruiter dashboard for score-ranked candidates and status updates, and a login-free candidate portal to track application status by ID and email.
.other projects
.say hello
Hiring for a backend role? Drop me an email and let's talk.