LOADING FAIRSWARM
FairSwarm deploys 4 free AI models simultaneously. Each model audits your datasets from a distinct lens, then converges into one trusted FairSwarm Score. More reliable than any single model. Built for everyone.
0
Fairness Metrics
0
AI Models in Parallel
< 0
< 5 min Audit Time
COMPAS Dataset
Racial Bias Detected
Live audits running on FairSwarm (simulated)
COMPAS Dataset · Racial Bias · Score 76.4 · Grade D · CRITICAL
Adult Income · Gender Bias · Score 61.2 · Grade C · HIGH
Hiring Algorithm · Age Bias · Score 43.1 · Grade B · MEDIUM
Credit Scoring · Regional Bias · Score 28.7 · Grade A · LOW
Medical Triage AI · Disability Bias · Score 81.3 · Grade F · CRITICAL
COMPAS Dataset · Racial Bias · Score 76.4 · Grade D · CRITICAL
Adult Income · Gender Bias · Score 61.2 · Grade C · HIGH
Hiring Algorithm · Age Bias · Score 43.1 · Grade B · MEDIUM
Credit Scoring · Regional Bias · Score 28.7 · Grade A · LOW
Medical Triage AI · Disability Bias · Score 81.3 · Grade F · CRITICAL
The Problem
Algorithms now decide who gets a job, loan, bail, or treatment. Flawed historical data silently encodes decades of discrimination.
0%
of hiring AIs show gender bias (MIT study)
0.0x
more likely Black defendants flagged high-risk by COMPAS
0%
of medical AI systems underperform on minority populations (NEJM)
Bias in AI is often invisible. Models learn from historically biased data and replicate discrimination silently.
A biased human hurts one person. A biased algorithm can hurt millions each day at machine speed.
Most teams lack practical, explainable bias auditing and compliance-grade evidence.
FairSwarm changes this. Upload your dataset or model and instantly see what is biased, why it matters, and how to fix it.
See How It WorksHow It Works
Upload your dataset (CSV, Excel, JSON) or connect your ML model. FairSwarm auto-detects sensitive attributes like gender, race, age, and religion.
4 AI models fire simultaneously. Each model acts as a specialist in statistical, contextual, historical, and intersectional bias.
Results are aggregated via weighted voting. The FairSwarm consensus is more accurate and harder to fool than any single model.
Download a professional PDF audit report with metric values, severity ratings, and prioritized remediation guidance.
The Core Innovation
Inspired by swarm intelligence, each FairSwarm agent is a specialist. Together they form a stronger and more trustworthy verdict.
FairSwarm
Consensus
Single models have blind spots. FairSwarm ensures one model missing a bias pattern does not hide critical risk. Weighted aggregation prevents any agent from dominating the final verdict.
If one agent fails or times out, the swarm continues with remaining agents. FairSwarm degrades gracefully.
Features
CSV, Excel, JSON support up to 50MB. Auto-detects columns, sensitive attributes, and target variable with zero configuration.
Disparate Impact Ratio, Statistical Parity, Equal Opportunity, Average Odds, Theil Index, Demographic Parity, Predictive Parity.
4 AI models analyze simultaneously. Weighted consensus yields one authoritative FairSwarm Score (0–100).
Detect bias at intersections of gender, race, and age with an intuitive heatmap view built for fast triage.
Professional 15-page reports with explanations, metric tables, and ranked remediation actions.
Track trends over time and compare before/after audits as you iterate on your model pipeline.
The Swarm
FairSwarm uses freely available AI APIs to make enterprise-grade bias detection accessible to every team.
Agent 1 · Statistical Analyst
NVIDIA NIM
Statistical Bias AnalysisAnalyzes disparate impact ratios, threshold violations, and statistical significance across demographic slices.
$25 free credits on signup
30% of final score
Get keyAgent 2 · Contextual Expert
Google AI Studio
Contextual Fairness AnalysisEvaluates social context and real-world impact on historically marginalized communities.
Free forever · 15 req/min
30% of final score
Get keyAgent 3 · Speed Specialist
Groq Cloud
Historical Pattern DetectionMaps findings against known bias patterns in lending, hiring, justice, and healthcare history.
Free · 30 req/min · Fast inference
25% of final score
Get keyAgent 4 · Intersection Detector
HuggingFace
Intersectional Bias DetectionDetects hidden risk at intersections of race, gender, age, and other protected attributes.
Free Inference API
15% of final score
Get keyLive Demo
COMPAS is a real algorithm used by US courts. FairSwarm reproduces known racial bias signals automatically.
FairSwarm Analysis Engine · COMPAS Dataset
> Uploading compas_scores.csv (500 rows)...
This is a real dataset with real bias. FairSwarm found it in 4.2 seconds.
Audit your own dataFairness Metrics
Disparate Impact compares selection rates between privileged and unprivileged groups. Values below 0.80 often indicate adverse impact. FairSwarm flags severity and recommends balancing strategies like reweighing.
Built With
Next.js 15.1 · TypeScript 5.7 · Tailwind CSS 3.4 · Framer Motion 11
FastAPI 0.115 · Python 3.12 · Pydantic v2 · Uvicorn
Supabase PostgreSQL · Auth · Storage · Row Level Security
IBM AIF360 · Microsoft Fairlearn · scikit-learn · pandas 2.2
NVIDIA NIM · Google Gemini · Groq Cloud · HuggingFace
ReportLab 4.2 · D3.js 7 · Recharts 2.13 · Python qrcode
Vercel · Render · GitHub Actions
JWT Auth · Rate Limiting · AES-256 · OWASP top-10
All services on free tier — built for students, researchers, and startups.
UN SDG Alignment
FairSwarm directly advances SDG 10, SDG 16, and SDG 8.
FairSwarm helps organizations detect and reduce discrimination before models affect real lives.
Transparent bias audits create accountability and defensible governance for AI-assisted decisions.
Fair hiring systems protect equal opportunity and prevent discriminatory rejection at scale.
India’s AI adoption across banking, government welfare, and healthcare is accelerating. Without robust bias auditing, these systems can encode caste, gender, and regional inequities at national scale.
340M+ Indians may be affected by AI decisions in hiring and lending by 2028 (NASSCOM)
Documentation
Base URL: https://fairswarm-backend.onrender.com/api/v1
Auth: Bearer JWT token
POST /datasets/upload Upload dataset file
POST /analysis/start Start bias analysis
GET /analysis/{id} Get results
GET /reports/{id}/pdf Download PDF report
POST /ai/triage Run swarm analysisOpen API Docs Includes Executive Summary, Dataset Overview, 7 Metrics, Consensus, Heatmap, Recommendations, Methodology, and Glossary.
Export formats: PDF and JSON
Sample ReportWhat People Say
Illustrative quotes for demonstration purposes
“FairSwarm found gender bias in our hiring algorithm in 4 minutes. The plain-English report made it easy to align with HR leadership.”
Anjali Sharma
Data Scientist, Bangalore Tech Startup
“When I tested COMPAS, FairSwarm surfaced the same racial bias pattern ProPublica documented. The swarm consensus is remarkably reliable.”
Dr. Priya Nair
ML Researcher, IIT Bombay
“As a compliance officer, the PDF audit report is exactly what I need for board review and regulatory conversations.”
Vikram Mehta
AI Governance Lead, HDFC Digital
FAQ
Yes. FairSwarm uses free-tier infrastructure and free AI API tiers, making it accessible for students, researchers, and early-stage teams.
Contact
Questions about FairSwarm, collaboration opportunities, or bias auditing for your organization?
GitHub Repository
github.com/Ankitkr-ak007/fairswarmLive Demo
fairswarm.vercel.appLocation
India · Google Solution Challenge 2026
Documentation
Full docs availableOpen Source
MIT License · Free forever