Prismo Finance

Inspiration

Managing personal finances in Malaysia is uniquely challenging. From navigating complex tax relief categories (SSPN, lifestyle expenses, medical) to tracking EPF contributions and multiple credit cards, we saw an opportunity to build something smarter. We were inspired by the question: What if your financial app could actually understand and advise you, not just display numbers?

The rise of conversational AI made us realize that personal finance doesn't have to feel like accounting, it can feel like talking to a knowledgeable friend who knows your complete financial picture.

What it does

Prismo Finance is an AI-native personal finance platform that transforms how Malaysians manage money:

• Conversational Financial AI: Chat naturally about your finances—"How much did I spend on food this month?" or "Am I on track for my savings goals?"
• Smart Tax Relief Tracking: Automatically categorizes expenses into Malaysian tax relief categories (lifestyle, medical, education, SSPN) with real-time YoY comparisons
• Financial Health Score: A dynamic 0-100 score analyzing savings rate, budget adherence, debt utilization, and goal progress
• Intelligent Insights: AI-generated alerts for overspending, upcoming bills, and optimization opportunities
• Multi-provider AI: Supports OpenAI, Anthropic Claude, and Azure OpenAI for flexibility and reliability

How we built it

Tech Stack:

• Frontend: Next.js 16, React 19, TailwindCSS, Framer Motion, Recharts
• Backend: Next.js API Routes, Drizzle ORM, PostgreSQL
• AI Architecture:
  • Agentic RAG for context-aware financial queries
  • CRAG (Corrective RAG) with hallucination checking and query rewriting
  • Intent Classification to route queries to specialized retrievers (transactions, tax, income, credit cards)
• Auth: Stack Auth with Google OAuth
• State Management: TanStack Query for real-time data synchronization

AI Pipeline: User Query → Intent Classifier → Specialized Retriever → Relevance Grader → Response Generator → Hallucination Checker

Challenges we ran into

1. Malaysian Tax Complexity: Encoding 20+ tax relief categories with varying limits, eligibility rules, and documentation requirements
2. Real-time Financial Health: Computing accurate health scores across multiple financial dimensions without impacting performance
3. AI Hallucination Prevention: Financial data demands accuracy—we implemented CRAG with relevance grading to ensure AI responses are grounded in actual user data
4. Multi-currency & Localization: Supporting MYR formatting, Malaysian date conventions, and local bank integrations

Accomplishments that we're proud of

• Sub-second AI responses through optimized retrieval pipelines
• Comprehensive tax relief automation covering all major Malaysian categories
• Customizable dashboard with drag-and-drop card reordering that persists to user accounts
• Zero-config AI switching between OpenAI, Claude, and Azure based on user preference or availability

What we learned

• RAG isn't enough: Corrective RAG with hallucination checking is essential for financial applications where accuracy is non-negotiable
• Domain-specific retrievers outperform generic ones: Separate retrievers for tax, transactions, and credit cards dramatically improved response quality
• Financial UX requires progressive disclosure: Users want quick insights upfront with the ability to drill deeper

What's next for Prismo Finance

• Bank Integration: Direct connections to Malaysian banks (Maybank, CIMB, Public Bank) via open banking APIs
• Predictive Forecasting: AI-powered cash flow predictions and expense forecasting
• Tax Filing Export: One-click export to LHDN e-Filing format
• Family Finance: Shared household budgets with role-based permissions
• Voice Interface: "Hey Prismo, how's my budget looking?"

Built With

• appwrite
• claude
• cursor
• gpt-5.1
• neonauth
• next.js
• openai
• opus
• opus4.5
• postgresql
• recharts
• stackauth
• tailwind
• vercel