BuckBounty

Inspiration

We noticed a common problem: people leave thousands of dollars on the table every year through suboptimal credit card usage, missed deals, and lack of financial awareness. Traditional budgeting apps just show you numbers, but they don't actively help you make money. We wanted to build an AI-powered financial copilot that doesn't just track spending—it turns every transaction into an opportunity to save, optimize, and build wealth.

What it does

BuckBounty is an AI-powered personal finance assistant that:

• Maximizes savings through credit card reward optimization (analyzes your spending patterns and recommends the best cards for each category)
• Finds hidden deals via automated coupon scraping from Gmail, Honey, and Rakuten
• Builds wealth by converting your savings into personalized investment portfolios with 10-year projections
• Provides intelligent insights using a multi-agent AI system (MARK orchestrates specialized agents for coupons, financial news, and market analysis)
• Offers voice-enabled chat with ElevenLabs text-to-speech for hands-free financial management
• Integrates with Plaid for automatic transaction syncing and real-time spending analysis
• Analyzes prediction markets through PolyMarket integration with risk assessment and ROI calculations
• Features @ mentions (e.g., @Starbucks) to instantly analyze spending at specific merchants and find relevant coupons

How we built it

Frontend:

• Next.js 14 with TypeScript for a modern, responsive interface
• ShadCN UI components with custom glass morphism effects
• Recharts for interactive spending visualizations (radar charts, pie charts)
• Vanta.js for immersive 3D animated backgrounds
• Web Speech API + ElevenLabs for voice interactions

Backend:

• FastAPI server with multiple specialized AI agents using MCP (Multi-agent Coordination Protocol)
• Google Gemini for natural language processing and embeddings
• RAG (Retrieval-Augmented Generation) with dual vector indexing:
  • FAISS FLAT index for current month (fast exact search)
  • HNSW index for historical data (efficient approximate search)
• Redis caching layer for sub-50ms response times on repeated queries
• Web scraping agents (BeautifulSoup + Selenium) for automated coupon aggregation
• Plaid API for secure banking integration
• Stripe for bill splitting functionality

Key Innovation: Multi-agent architecture where MARK (main agent) orchestrates specialized BountyHunter agents—one scrapes deals, another fetches personalized financial news based on your spending categories.

Challenges we ran into

1. Vector search scalability: Initially, semantic search on 10,000+ transactions was slow. We solved this by implementing a dual-index system—FLAT for recent data, HNSW for historical—with automatic migration.

2. LLM response caching: Gemini calls were expensive and slow (~2.5s per request). We built a Redis-based caching layer that reduced identical queries to <50ms, a 50x improvement.

3. Merchant name matching: User transactions have messy merchant names (e.g., "STARBUCKS #12345 SEATTLE WA"). We implemented fuzzy matching with embeddings to accurately match @Starbucks mentions to actual transaction data.

4. Real-time agent coordination: Coordinating multiple AI agents (MARK, BountyHunter1, BountyHunter2) without conflicts required building a custom MCP server with status tracking and task queuing.

5. Credit card optimization logic: Calculating optimal rewards across categories with annual fees was complex. We built a comprehensive credit card database with reward structures and developed an algorithm that maximizes net savings after fees.

Accomplishments that we're proud of

• 50x performance improvement through intelligent Redis caching
• Multi-agent AI orchestration that actually works in production
• Dual-index RAG system handling 10,000+ transactions with sub-second search
• Automated wealth building that converts credit card savings into actionable investment portfolios with real fund recommendations
• Beautiful UI/UX with glass morphism, 3D backgrounds, and smooth animations
• Voice-enabled interface that makes finance management accessible hands-free
• @ mention feature that provides instant merchant-specific insights
• Production-ready security with Plaid OAuth and proper API key management

What we learned

• RAG architecture: Implementing FAISS with dual indexing taught us the tradeoffs between exact and approximate search. FLAT is perfect for small, current datasets while HNSW scales to millions of vectors.
• LLM optimization: Caching isn't just about speed—it saves real money. Our Redis layer reduced API costs by 70%.
• Multi-agent systems: Coordination is hard. We learned to design agents with clear responsibilities and avoid circular dependencies.
• Financial domain knowledge: Credit card reward structures, investment portfolio theory, and prediction markets—building BuckBounty required deep research into personal finance.
• UX matters: The most powerful AI is useless if the interface is confusing. We iterated heavily on the dashboard and chat experience.

What's next for BuckBounty

1. Machine learning for budget predictions: Train models on transaction history to predict future spending and warn users before they overspend.

2. Tax optimization agent: Automatically identify tax deductions from transactions (charitable donations, business expenses, medical costs).

3. Automated bill negotiation: Use AI to call service providers (phone, internet, insurance) and negotiate lower rates.

4. Social features: Allow users to compare anonymized spending patterns with peers and compete on savings leaderboards.

5. Mobile app: React Native app with push notifications for real-time deal alerts and bill reminders.

6. Open banking expansion: Support for international banks beyond Plaid's coverage using Teller, Finicity, and MX.

7. Cryptocurrency integration: Track crypto portfolios alongside traditional assets and optimize tax-loss harvesting.

8. AI-powered financial education: Personalized learning modules based on user's financial weaknesses (e.g., if overspending on dining, provide meal prep strategies).

BuckBounty's mission: Make financial optimization accessible to everyone, not just those who can afford financial advisors. Every dollar should work for you, not against you.

Built With

• beautiful-soup
• elevenlabs-api
• faiss-vector-database
• fastapi
• google-gemini-api
• next.js-14
• plaid-api
• polymarket-api
• python
• react-18
• recharts
• redis
• selenium
• sentence-transformers
• shadcn-ui
• stripe
• tailwind-css
• typescript
• vanta.js
• web-speech-api
• yahoo-finance-api
• zustand