@ Mohamed Amine El Abidi

ARYA multi-agent recruitment workflow

AI agent engineer, Valhko

ARYA: multi-agent recruitment workflow. Illustrated project cover.

Illustrated project cover.

Source-backed explainer

ARYA in 20 seconds

Public backend snapshot · commit fbb10ee

A six-agent recruitment workflow covering job descriptions, skill extraction, candidate assignments, submission evaluation, scoring and ranking, orchestrated with LangGraph and exposed through FastAPI.

ARYA is a six-agent recruitment workflow covering job descriptions, skill extraction, candidate assignments, submission evaluation, scoring and ranking. LangGraph coordinates the hand-offs, while FastAPI exposes the workflow through explicit request and response contracts.

Azure OpenAI provides model calls, SQLAlchemy persists jobs, candidates, projects and submissions, and the public backend exposes the evaluation and ranking flow. Deployment artifacts document Docker and Azure App Service delivery.

The public repository makes the backend contracts inspectable: endpoints, Pydantic schemas, evaluation services, scoring logic, Docker build and CI workflow. Source code is public on GitHub.

Inspectable evidence

A recruitment workflow with visible contracts and scoring.

The public repository exposes the API, models, evaluation services, container build and CI workflow. The product context came from Valhko; the repository is an inspectable backend snapshot, not a hosted production demo.

Workflow

  1. 01

    Job

    Create requirements and an assessment project.

  2. 02

    Candidate

    Register a candidate and evaluate CV text against the job.

  3. 03

    Submission

    Accept one submission per project phase and persist the evaluation.

  4. 04

    Ranking

    Combine CV and project scores, then return an ordered candidate list.

Published ARYA Azure architecture diagram showing API, agent and service layers.
Architecture diagram published in the public ARYA repository. It documents the intended Azure deployment boundaries.

What the source demonstrates

API surface

FastAPI endpoints for jobs, candidates and submissions, with Pydantic request/response schemas.

Evaluation logic

The public service code evaluates CVs and phased submissions, prevents duplicate phase submissions and persists results.

Transparent scoring

When project submissions exist, the public ranking function weights CV evidence at 30% and project evidence at 70%.

Delivery

Dockerfile, GitHub Actions import checks and Docker builds, deployment documentation and architecture diagrams are published with the code.

Other projects

  • Netix AI network assistant (Orange)
  • CasaMotion real-time mobility data platform
  • DocAssist RAG over real-estate documents
  • Replenish batch data pipeline for retail procurement
  • Radian job intelligence warehouse