login page
data visualization
audit dashboard
email generation to employee highliting their transactions
main dashboard

SUSpend - Local AI Financial Auditing Dashboard

Inspiration

The financial sector increasingly relies on third-party cloud services to handle sensitive compliance and auditing tasks. Through conversations with industry professionals, we discovered that internal auditing and corporate compliance teams regularly send private transactional data to external cloud providers—creating unnecessary security risks and privacy vulnerabilities.

We recognized a critical opportunity: by leveraging local AI inference, we could deliver enterprise-grade auditing capabilities while keeping sensitive financial data completely on-premise. This approach eliminates the compliance overhead and data exposure that companies currently face with cloud-based solutions.

What It Does

SUSpend is a local AI-powered dashboard for internal auditors that provides real-time compliance monitoring of employee transactions. The system:

Analyzes employee transactions directly from connected bank accounts
Validates spending patterns against customizable company policies in real-time
Flags suspicious activity using locally-running AI models—no cloud data transmission
Manages audit tickets where employees can raise concerns and auditors can respond
Maintains data sovereignty by processing all sensitive financial information locally

How We Built It

Core Architecture:

AI Engine: Local AI inference via Ollama for policy validation and anomaly detection
Frontend: React and Next.js for an intuitive auditor dashboard
Backend: Python Flask API serving the AI models and managing data flows
Data Pipeline: Seamless integration between transaction data, policy validation, and compliance reporting

Development Process: We organized into domain-focused teams leveraging individual strengths—AI/backend, frontend, and integration—with disciplined version control to ensure smooth coordination and timely delivery.

Challenges We Overcame

AI Model Consistency: Large language models occasionally hallucinate or generate inconsistent outputs when parsing financial data. We implemented validation layers and prompt engineering to improve accuracy and reliability.

Cross-Domain Integration: Coordinating between independently developed AI models, Flask backend, and React frontend required careful API contracts and extensive testing to ensure seamless data flow.

Real-Time Processing: Optimizing local inference speed to provide instant feedback on transaction auditing without server infrastructure.

Accomplishments We're Proud Of

✓ Successfully ran complex background AI processes smoothly on standard personal computers ✓ Built a fully functional end-to-end system from data intake through policy validation to user interface ✓ Demonstrated that enterprise-grade compliance tooling can operate entirely on-premise ✓ Created a scalable foundation that would perform significantly better on dedicated company servers

What We Learned

Local AI inference is production-viable for sensitive financial applications
Careful prompt engineering is critical for consistent financial data analysis
Privacy-first architecture resonates strongly with enterprise requirements
Domain-driven team organization accelerates complex full-stack development

What's Next for SUSpend

RIT Campus Finance Mode - Specialized deployment for RIT student organizations, automating budget compliance and club fund auditing with policy rules tailored to university governance.

Insurance Claims Validation - Extended the auditing engine to compare insurance claim submissions against actual transaction records, detecting discrepancies locally before submission to insurance providers.

Enhanced Policy Engine - Support for dynamic, role-based policy definitions and automated policy recommendation based on industry standards.