RecruitLens

Know what the algorithm sees.

ATS simulation · Algorithmic bias detection · Research-backed

---

What is RecruitLens?

RecruitLens is an AI-powered tool that shows job seekers exactly how an Applicant Tracking System (ATS) processes their resume — and flags resume elements that peer-reviewed research identifies as proxies for protected characteristics.

Most ATS systems are black boxes. Candidates submit resumes without knowing whether they'll survive automated keyword filtering, and without knowing that completely neutral elements — their name, their address, their graduation year — can cause an algorithm to silently filter them out before a human ever reads their application.

RecruitLens makes that process transparent.

Upload a resume and a job description. In seconds, you get:

• An ATS score (0–100) simulating keyword and skills matching
• A pass / fail prediction with plain-English reasoning
• Extracted entities — what an ATS actually sees: skills, education, experience
• Keyword gaps — hard skills, soft skills, and certifications missing from your resume
• Bias signal detection — 8 research-backed proxy discrimination patterns flagged with citations
• Actionable suggestions — specific rewrites, not generic advice

---

Screenshots

---

The Research

This tool is the practical implementation of a peer-reviewed study published in REVIZOR — Journal for Organization Management, Finance and Audit.

Pavlović, M., Glišović, L., & Mirčetić, V. (2026). Uticaj veštačke inteligencije na savremene prakse regrutacije i selekcije [The Impact of Artificial Intelligence on Contemporary Recruitment and Selection Practices]. REVIZOR, 28(4). https://doi.org/10.46793/Rev25112.091P

The paper documents how AI recruitment systems perpetuate historical discrimination through proxy bias — using seemingly neutral data points as stand-ins for protected characteristics. It identifies 8 specific patterns, each grounded in independent algorithmic bias research:

Signal	Proxies for	Severity	Source
Graduation year	Age	Medium	Köchling & Wehner (2020)
Foreign name	Ethnicity / national origin	High	Mehrabi et al. (2021)
Home address	Socioeconomic status	Low	Raghavan et al. (2020)
Employment gap	Disability or caregiving	Medium	Glazko et al. (2024)
Disability language	Disability status	High	Glazko et al. (2024)
Photo reference	Gender, ethnicity, age	High	Bogen & Rieke (2018)
University prestige	Socioeconomic class	Low	Raghavan et al. (2020)
Gendered language	Gender	Medium	Köchling & Wehner (2020)

RecruitLens detects all 8. Every flagged result includes the triggering excerpt, a plain-English explanation, and the academic citation — making the analysis fully transparent and explainable.

---

Tech Stack

Layer	Technology
Frontend	React 18 · Vite 5 · React Router v6
Backend	Python 3.11 · FastAPI · Uvicorn
AI	Google Gemini 2.5 Flash
Document parsing	pdfplumber · python-docx
Security	Magic-byte file validation · CSP headers · CORS · Rate limiting
Hosting	Vercel (frontend) · Render (backend)

---

Getting Started

Prerequisites

• Node.js 18+
• Python 3.11+
• A Google AI Studio API key (free tier: 1,500 req/day)

1. Clone

git clone https://github.com/matejapp/recruit-lens.git
cd recruit-lens

2. Backend

cd server
pip install -r requirements.txt

Create server/.env:

GEMINI_API_KEY=your_api_key_here
MODEL_NAME=gemini-2.5-flash
ALLOWED_ORIGIN=http://localhost:5173

Start the server:

uvicorn main:app --reload
# API running at http://localhost:8000
# Swagger docs at http://localhost:8000/docs

3. Frontend

cd client
npm install

Create client/.env.local:

VITE_API_URL=http://localhost:8000

Start the dev server:

npm run dev
# App running at http://localhost:5173

---

Deployment

Frontend → Vercel

1. Import the repo on vercel.com
2. Set Root Directory to client
3. Add environment variable: VITE_API_URL = https://your-app.onrender.com
4. Deploy — Vercel auto-detects Vite

Backend → Render

The repo includes a render.yaml Blueprint. On render.com:

1. New → Blueprint → connect this repo
2. Set two environment variables manually in the dashboard:
   • GEMINI_API_KEY — your Google AI Studio key
   • ALLOWED_ORIGIN — your Vercel deployment URL

No Docker required — Render runs the FastAPI server natively.

Note: Render's free tier sleeps after 15 minutes of inactivity. The first request after sleep takes ~30 seconds. The frontend's 90-second timeout handles this gracefully.

---

Privacy

RecruitLens is stateless by design. No resume data, job description text, or analysis results are stored, logged, or persisted anywhere. Every request is fully isolated and processed in memory only.

---

Disclaimer

RecruitLens is decision support — not a hiring decision. All findings are based on peer-reviewed research on algorithmic bias in recruitment. Results should always be combined with human judgment.

---

License

MIT — see LICENSE.

---

_{Built on research published in REVIZOR Journal · Faculty of Applied Management, Economics and Finance · University Business Academy in Novi Sad}