CREDIS

The Problem

India has 450 million gig workers.

Swiggy delivery partners. Uber drivers. Urban Company professionals. They wake up early, work late, pay rent on time, save money, and support their families.

Banks still reject them.

Not because they are poor. Not because they are irresponsible. But because they don't have a salary slip.

Traditional credit scoring was built for salaried employees in the 1970s. It has never been updated for the gig economy. The result: 450 million financially invisible workers who cannot access loans, insurance, or formal financial services — despite having real, verifiable income.

The Insight

Gig workers are not financially invisible.

Their DATA is invisible.

Scattered across Swiggy payouts, UPI transactions, electricity bills, recurring deposits, and mobile recharges — there is more than enough behavioral signal to build a complete financial identity for any gig worker.

The question is: who will collect it, analyze it, and turn it into a credit score that banks will trust?

CREDIS does exactly that.

What We Built

CREDIS is a behavioral credit intelligence system that scores gig workers using 6 autonomous ASI-1 powered agents.

Each agent has a specific role in the pipeline:

Agent 1 — DataWeaver Connects to the worker's bank via India's RBI-regulated Account Aggregator framework. Collects transaction history, RD installments, insurance records, and platform earnings — all with one-time digital consent. No documents. No uploads.

Agent 2 — PatternMind Analyzes 12 months of behavioral data. Calculates income volatility, active working months, rent payment consistency, and income growth trend between first and second half of year.

Agent 3 — IdentityForge Calculates the CREDIS Score on 5 weighted signals:

Income Stability (30%)
Platform Reliability (25%)
Work Consistency (20%)
Spending Discipline (15%)
Income Growth (10%)

Applies CKYC and NBCFDC adjustments for accuracy.

Agent 4 — RiskGuard Classifies the worker into one of four risk bands and generates a loan decision — Approve Fast, Approve Standard, Approve Capped, or Coaching Only.

Agent 5 — OpportunityScout Scans a database of 8 real Indian lenders including NBCFDC, PM SVANidhi, MUDRA, Ujjivan SFB, and Arohan MFI. Matches the worker to eligible lenders ranked by interest rate.

Agent 6 — GrowthCoach Generates a personalized 12-month financial roadmap — Foundation, Growth, Expansion, and Consolidation phases — with specific actions and year-end income targets.

Data Standard

Worker data follows the official ReBIT Account Aggregator schema from the Sahamati repository — the same standard used by HDFC, SBI, Kotak, and all AA-registered banks in India.

This means CREDIS is production-ready from day one. The data pipeline is not a prototype. It is the real standard.

Results

We scored 5 real worker profiles across Chennai, Madurai, Coimbatore, Bengaluru, and Hyderabad:

Worker	Platform	CREDIS Score	Decision
Ravi Kumar	Swiggy	98.8 / 100	Approve Fast
Meena Devi	Urban Company	91.2 / 100	Approve Fast
Arjun Das	Uber	60.2 / 100	Approve Standard
Priya Lakshmi	Zomato	100 / 100	Approve Fast
Suresh Babu	Swiggy	93.5 / 100	Approve Fast

What We Learned

Building CREDIS taught us that the hardest problem in financial inclusion is not technology — it is trust.

Banks don't trust gig workers because they can't verify their income. Gig workers don't trust banks because they have been rejected too many times.

CREDIS sits in the middle. It uses technology — ASI-1 agents, Account Aggregator, behavioral analytics — to build a bridge between the two.

We also learned that India already has the infrastructure for this. The Account Aggregator framework is live. The lenders exist. The workers exist. What was missing was the intelligence layer. That is what CREDIS provides.

Challenges

Challenge 1 — Data Schema Designing worker JSON files that exactly match the ReBIT Account Aggregator standard required deep study of the Sahamati repository. Every field — transaction type, amount, date, category — had to match the official schema.

Challenge 2 — Agent Chaining Building a reliable 6-agent pipeline where each agent passes a clean Python dictionary to the next required careful state management. A single error in one agent cascades through the entire chain.

Challenge 3 — Scoring Formula Calibrating the 5-signal CREDIS Score so that workers with genuinely different financial profiles received meaningfully different scores — not just similar numbers — required multiple iterations of weight tuning.

What's Next

Phase 2 — Live Integration Connect React frontend to FastAPI backend so loan officers can enter a worker's AA consent ID and trigger a live agent pipeline run in real time.

Phase 3 — Mobile App Worker-facing mobile app where gig workers can connect their bank account, see their CREDIS Score, and apply for loans directly — in Tamil, Hindi, and Telugu.

Phase 4 — Lender API Direct API integration with NBCFDC, MUDRA, and PM SVANidhi so approved workers receive loan offers inside the app within minutes of scoring.