SahiScore AI

Inspiration

Over 350 million women in India remain unbanked — not because they lack financial discipline, but because traditional credit scoring systems like CIBIL require formal banking histories that these women simply don't have. Despite running thriving micro-businesses with a remarkable 98% repayment rate in Self-Help Groups (SHGs), they remain invisible to lenders.

We were inspired by the paradox: the most creditworthy borrowers in India are the ones who can't get a credit score.

What it does

SahiScore AI is an alternative credit scoring system that generates reliable credit scores (300-900) for unbanked women entrepreneurs using non-traditional data sources:

• 💳 Financial Behavior (35%) — Utility bill payments, mobile recharge patterns, SHG savings contributions, internal loan repayment history
• 📊 Economic Activity (30%) — Years in business, monthly revenue, supply chain participation, product diversity, marketplace activity
• 🤝 Community Trust (25%) — SHG meeting attendance, peer trust scores, community references, leadership roles
• 🏠 Stability (10%) — Residential stability, SHG membership duration, family situation

The system provides:

• Explainable AI — Every score comes with a SHAP-inspired factor contribution chart showing exactly why the applicant received their score
• Bias-Free Scoring — Gender, caste, religion, ethnicity, and location are explicitly excluded from all calculations
• Loan Eligibility — Automatic recommendations for loan amount, interest rate, and tenure
• MFI Dashboard — A comprehensive overview for micro-finance officers to manage all assessments

How we built it

• Frontend: Vite + vanilla JavaScript with a custom CSS design system featuring dark theme, glassmorphism, and micro-animations
• Scoring Engine: A weighted multi-factor model inspired by real ML credit scoring approaches, with explainability output modeled after SHAP (SHapley Additive exPlanations)
• Visualizations: Chart.js for interactive factor contribution charts, score distribution histograms, and risk breakdown doughnuts
• Demo System: 3 pre-built personas representing real-world archetypes (experienced weaver, new entrepreneur, seasonal farmer) for instant walkthroughs

Challenges we ran into

• Designing a scoring model that is both realistic and demonstrably fair — we had to carefully choose which alternative data points to include while ensuring none could serve as proxies for protected characteristics
• Balancing explainability with simplicity — the SHAP-style factor chart needed to be understandable by field workers with limited technical background
• Creating a mobile-friendly interface that works in low-connectivity rural environments while still looking premium

Accomplishments that we're proud of

• Built a fully functional prototype with 4 complete pages and a working scoring engine
• The system assessed 9 demo applicants, with 8 qualifying for micro-loans — unlocking an estimated ₹4.65 lakh in potential credit
• Every single score is 100% explainable — no black-box decisions
• Zero bias by design — the model explicitly excludes all protected characteristics

What we learned

• Alternative data is surprisingly predictive: SHG attendance and peer trust scores correlate strongly with repayment behavior
• Explainability isn't just an ethical requirement — it's a trust-building tool that makes both lenders and borrowers more confident
• Financial inclusion technology needs to be designed for the field, not for the boardroom

What's next for SahiScore AI

• Partner with actual SHG federations for real-world pilot testing
• Integrate with Aadhaar-based e-KYC for identity verification
• Add offline-first PWA capabilities for areas with limited connectivity
• Train a proper ML model on anonymized SHG repayment data
• Build a regional language interface (Hindi, Tamil, Bengali) for direct applicant interaction

Built With

• chart-js
• css
• html
• javascript
• machine-learning
• vite