Context
Hiring teams needed a backend that could store resumes, pull jobs, and score matches instead of reading every PDF by hand.
AI & Agentic Systems
Hiredesk - Recruiting Match System
A recruiting match desk that parses resumes, ingests jobs, and ranks fit.
Software build · AI Recruiting Platform · FastAPI + Celery + Postgres + Redis + embeddings
Project illustration. Not a verified production screenshot.
Engagement: Software build. The case study documents the implementation scope. Live deployment status and usage are not asserted.
The capabilities below describe the work delivered. Quantified business results have not been published.
Challenge
Matching quality depends on parsing, embeddings, and a second-pass rerank. Job ingest also has to run on a worker, not on the request path.
Solution
Avlys built a FastAPI service with Celery workers, Postgres models for resumes, jobs, and matches, JSearch ingest, embedding similarity, and LLM reranking.
Inside the project
System architecture.
Logical components and integration boundaries. This diagram is not a deployment or compliance certification.
Architecture explained in text
The web client calls FastAPI with Supabase-backed authentication. Application modules parse resumes, collect jobs from JSearch, perform vector and keyword matching, and rerank matches through an OpenAI-compatible API. Postgres with pgvector stores profiles, jobs, embeddings, and applications. Celery workers and beat run scheduled ingestion and embedding tasks, with Redis as broker and result backend. Provider credentials are required for the relevant features.
Delivered.
Resume parsing
Job ingest worker
Embedding match layer
LLM rerank and match APIs
Capabilities delivered.
Resumes become structured profiles instead of unread PDFs.
Jobs can be collected on a schedule instead of pasted one by one.
First-pass ranking is vector-based, with a slower LLM rerank on top.
A backend ready for a recruiter UI without rewriting the match core.
How it was built.
01
Defined profiles, jobs, and applications in Postgres.
02
Added scheduled ingestion and embedding tasks through Celery and Redis.
03
Combined vector and keyword retrieval with model-backed reranking.
04
Documented provider configuration, migrations, workers, and health checks.
Related work.
All case studies →
Mike Legal AI
A legal document workspace for ingest, conversion, and model-backed review.
Hospital Voice Agent
A trilingual hospital receptionist that books appointments from inbound calls.
BidSure AI - Tender Eligibility Platform
A tender desk that checks contractor eligibility before a bid is assembled.
Let’s scope your next system.
30 minutes, engineers on the call, no deck. We’ll map your workflow and tell you honestly whether AI pays back - and what a fixed-price pilot would look like.
Tell us about the project
Request a call or share your requirements. We reply within one business day.