When our team won PayHack 2024, Malaysia's premier Open Finance hackathon, it wasn't just about building another fintech app. It was about solving a real problem that affects millions of Malaysians.
As the Technical Lead for DuitWise, I want to share our journey of building a solution that bridges the gap between traditional banking and modern financial needs while tackling the critical challenges of financial literacy and inclusion in Malaysia.
The Challenge: Why We Built DuitWise
The numbers were stark:
- 64% of Malaysian adults are financially illiterate
- 49.22% of bankruptcy cases are due to financial loans
- 6.1 out of 10 Malaysians struggle to access bank loans
- 48.4% still prefer cash transactions (rising to 63.5% in rural areas)
We saw an opportunity to create something that wasn't just another banking app, but a comprehensive financial companion that could help users like Mr. Hazwan, a Grab rider, and Mei Lin, a semi-urban farmer to make better financial decisions.
Technical Architecture
At its core, DuitWise is built on three main pillars:
1. Backend Infrastructure
- Client-Side Processing: We implemented a unique approach where computationally intensive tasks run on the mobile device, reducing server costs and improving response times
- Mistral 7B Integration: A lightweight LLM for personalized financial advice
- OCR Donut Transformer: For receipt processing and categorization
- Python Scripts: For financial calculations and stress testing
2. Open Banking Integration
We plan to integrate with major Malaysian banks through their Open APIs:
- Maybank M2U Pay
- CIMB API Gateway
- Bank Islam Open Banking
- GX Bank API
3. Frontend Development
React Native: For cross-platform mobile development
Tailwind CSS: For consistent and responsive UI
Custom Components: Built reusable components for:
- Multi-bank dashboard
- Financial fitness scoring
- Loan stress testing
- Digital receipt management
Key Technical Features
1. ResitNow - Intelligent Receipt Management
- Computer Vision model for receipt content extraction
- Real-time tax relief and subsidy eligibility mapping
- Automated categorization using Python backend
2. Financial Stress Testing
- Real-time DSR (Debt Service Ratio) calculation
- Integration with multiple bank APIs for comprehensive assessment
- Machine learning model for personalized recommendations
Core Formula: Debt Service Ratio (DSR)
The foundation of our loan stress testing is an enhanced version of the traditional Debt Service Ratio:
DSR = (New_Monthly_Commitment / Total_Income) × 100
where:
New_Monthly_Commitment = Current_Total_Loan + Current_Total_Commitment + New_Monthly_Payment
However, we knew this basic formula wasn't enough for gig workers. Here's how we enhanced it:
For gig workers, we implemented a weighted average system:
Income Stability Assessment
Weighted_Monthly_Income = (3M_avg × 0.5) + (6M_avg × 0.3) + (12M_avg × 0.2)
where:
3M_avg = Average income for last 3 months
6M_avg = Average income for last 6 months
12M_avg = Average income for last 12 months
3. DuitSmart AI Agent
- Locally-run LLM for personalized financial advice
- Real-time transaction analysis
- Proactive investment suggestions based on user profile
The magic happens in how we process financial data. Here's a peek at our scoring algorithm:
Financial_Fitness = (0.3 × SR) + (0.25 × DM) + (0.2 × EFR) + (0.15 × IU) + (0.1 × ID)
Engineering Challenges and Solutions
1. Scale and Performance
To handle our projected user base (from 12k to 120k users), we implemented:
- Client-side processing for CPU-intensive tasks
- Efficient data storage (2MB per user)
- Optimized API calls using caching
2. Security and Compliance
End-to-end encryption for financial data
Compliance with Bank Negara Malaysia regulations
Secure multi-bank authentication
3. User Experience
- Offline-first architecture for rural users
- Progressive enhancement for varying device capabilities
- Sub-second response times for critical features
Impact and Results
Our solution has the potential to:
- Increase financial literacy by 35%
- Streamline targeted subsidy claims
- Empower 4 million gig workers with loan approval tools
- Help 15 million taxpayers with documentation
Conclusion
DuitWise demonstrates how technology can solve real-world problems at scale. By combining AI, open banking, and user-centric design, we've created a solution that makes financial management accessible to all Malaysians.
The Team Behind the Tech
As Technical Lead, I worked with an amazing team:
This project was developed for PayHack 2024 by Team Barracuda. I served as the Technical Lead, focusing on architecture and implementation.
Top comments (0)