Credence

What happens to you if you don't use Credence

Inspiration

Credence is a student-focused multi-LLM agent system built in response to a broken reality: students are expected to prove financial trust before they have had the chance to build traditional credit.

Millions of responsible students are labeled high risk simply because credit systems measure debt history, not financial behavior. Savings discipline, income consistency, and payment reliability are ignored, even though these signals better reflect real-world trust.

We asked ourselves:

What if financial trust were evaluated the way humans do, by patterns of behavior, not a single number?

Credence was created to answer that question.

What it does

Credence is a student-focused multi-LLM agent platform that evaluates financial trust using behavioral intelligence rather than legacy credit scores.

Instead of relying on a single opaque model, Credence deploys multiple specialized AI agents, each responsible for analyzing one dimension of a student’s financial life. Together, these agents generate a transparent, explainable Trust Score and actionable guidance.

At a high level:

Trust Score = Σ (wᵢ × Bᵢ)

Where:

Bᵢ represents a behavioral signal, including payments, savings, income stability, spending, and investments
wᵢ is an AI-derived weight adjusted per user context

Credence enables students to:

Understand their financial health in plain language
Manage tuition, loans, scholarships, and income in one place
Receive AI-driven financial strategies
Generate a Trust Passport, a privacy-preserving proof of reliability for landlords, employers, or lenders
Simulate future decisions and see how behavior affects trust over time

How we built it

Credence was designed as a multi-agent AI system, not a single monolithic model.

Multi-LLM Agent Architecture

Each AI agent specializes in a different financial dimension:

Consensus Score = (A₁ + A₂ + ... + Aₙ) / n

Payment Agent, payment reliability
Savings Agent, emergency fund and consistency
Income Agent, stability and risk
Spending Agent, behavioral patterns
Investment Agent, diversification and growth
Risk Agent, overall financial exposure

Agents independently analyze user data, then converge on a unified, explainable output.

Tech Stack

Frontend: React 18, Vite, Recharts, Context API
Backend: Node.js, Express
AI Layer: Python with Gemini API using MCP
Design: Glassmorphism, dark mode, responsive layouts

Challenges we ran into

Coordinating multiple AI agents Ensuring agents disagreed productively without producing contradictory outputs required careful normalization and consensus logic.
Privacy vs. verification Designing the Trust Passport meant proving reliability without exposing raw financial data.
Student UX clarity Financial systems are intimidating. We focused on clarity, visuals, and language that feels supportive rather than punitive.
Scope under hackathon constraints Building both a production-grade UI and a functioning AI system pushed us to prioritize architecture and execution.