Inspiration

Small businesses struggle with cash flow management — 82% fail due to poor cash flow visibility. We built StackSpark to change that by combining AI intelligence with persistent agentic memory.

What it does

StackSpark is an AI Cash Flow Copilot that uses CockroachDB's distributed SQL and agentic memory to:

  • Predict cash flow 30-90 days ahead using ML models
  • Track real-time income/expenses across multiple bank accounts
  • Optimize payment schedules and invoice timing
  • Remember financial patterns across sessions with persistent memory

How we built it

  • Backend: Python + FastAPI with CockroachDB for distributed, resilient data storage
  • AI Engine: LLM-powered financial analysis with RAG pipeline for contextual insights
  • Agentic Memory: CockroachDB stores vector embeddings of financial patterns, enabling the AI to learn and improve recommendations over time
  • Frontend: React + TypeScript dashboard with real-time charts
  • Infrastructure: AWS (EC2, S3, Lambda) for deployment and data processing

Challenges we ran into

  • Building reliable agentic memory that persists financial context across sessions
  • Optimizing CockroachDB queries for real-time cash flow calculations at scale
  • Balancing AI model accuracy with response latency for live financial data

Accomplishments that we're proud of

  • Achieved sub-second prediction response times with 87% accuracy
  • Built a fully distributed system that handles node failures gracefully
  • Created an AI agent that genuinely learns from financial patterns

What we learned

CockroachDB's distributed SQL is perfect for fintech — strong consistency without sacrificing availability. Agentic memory transforms AI from a chatbot into a true financial advisor.

What's next for StackSpark — AI Cash Flow Copilot

  • Multi-currency support for international businesses
  • Banking API integrations (Plaid, Open Banking)
  • Mobile app for on-the-go financial insights

Built With

Share this project:

Updates

Submission history