Smart Treasury

Smart Treasury Agent is an AI copilot for corporate cash.It connects to banks, simulates treasury scenarios in Daytona, and uses Galileo to score risk and return, helping CFOs move money instantly.

Comment

Inspiration

I have a startup in payments, and we've noticed that corporate treasury teams still manage millions using spreadsheets. They struggle to predict liquidity, optimize idle cash, or simulate “what-if” scenarios in real time. I want to give CFOs an intelligent copilot, a system that thinks like a treasury analyst, acts instantly, and learns over time.

What it does

Smart Treasury Agent is an AI copilot for corporate cash. It connects to banking APIs, simulates treasury scenarios inside isolated Daytona sandboxes, and uses Galileo to score each strategy’s risk and return, helping CFOs decide where to move money, instantly.

How we built it

Frontend: Nextjs Backend: Node.js + Supabase DB Daytona: Python Microservice running with fastAPI

Challenges we ran into

Parallel Simulation Architecture: Daytona SDK wasn’t stable in Node.js, so we pivoted to a Python FastAPI microservice for async, parallel simulations. Real-time Coordination: Managing sync between Daytona sandboxes, Python, and the React frontend required a custom polling system with React Query.

Accomplishments that we're proud of

True Parallel Execution: Three Daytona sandboxes running real simulations concurrently. Advanced AI System: 4-agent workflow, debate mechanism, historical learning, and Galileo observability.

What we learned

Microservices + Daytona = Power: Python for simulation, Node for orchestration delivered high performance. Structured Agents Win: Multi-agent prompts need clear roles and output formats. Daytona Parallelism Rocks: True parallel sandboxing cut simulation time drastically. Learning Loops Work: Historical feedback + Galileo made AI improvement measurable. TypeScript Saves Time: Type safety prevented dozens of runtime errors.

What's next for Smart Treasury

Live Banking APIs: Integrate Plaid, Stripe Treasury, or SVB for real balances and transfers. Advanced Risk Models: Add Monte Carlo, VaR, and stress tests. Collaboration: Multi-user workflows with approval chains and audit trails. Mobile Dashboard: Real-time approvals and monitoring. AI Trading & FX: Expand to treasury securities and market-driven strategies. Enterprise Scale: Multi-entity support and regulatory compliance (SOX, SOC 2, Basel III).

Built With

Share this project:

Updates